CA2660202C - Detecting duplicate and near-duplicate files - Google Patents
Detecting duplicate and near-duplicate files Download PDFInfo
- Publication number
- CA2660202C CA2660202C CA2660202A CA2660202A CA2660202C CA 2660202 C CA2660202 C CA 2660202C CA 2660202 A CA2660202 A CA 2660202A CA 2660202 A CA2660202 A CA 2660202A CA 2660202 C CA2660202 C CA 2660202C
- Authority
- CA
- Canada
- Prior art keywords
- documents
- document
- duplicate
- computer
- implemented method
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/194—Calculation of difference between files
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Near duplicate documents may be identified by processing an accepted set of documents to determine a first set of near duplicate documents using a first technique, and processing the first set to determine a second set of near duplicate documents using a second technique. The first technique might be token order dependent, and the second technique might be order independent. The first technique might be token frequency independent, and the second technique might be frequency dependent. The first technique might determine whether two documents are near duplicates using representations based on a subset of the words or tokens of the documents, and the second technique might determine whether two documents are near duplicates using representations based on all of the words or tokens of the documents. The first technique might use set intersection to determine whether or not documents are near duplicates, and the second technique might use random projections to determine whether or not documents are near duplicates.
Description
DETECTING DUPLICATE AND NEAR-DUPLICATE FILES
1. BACKGROUND OF THE INVENTION
1.1 FIELD OF THE INVENTION
[0001 ] The present invention concerns information management and retrieval in general.
More specifically, the present invention concerns detecting, and optionally removing, duplicate and near-duplicate information or content, such as in a repository of documents to be searched for example.
1.2 BACKGROUND INFORMATION
1. BACKGROUND OF THE INVENTION
1.1 FIELD OF THE INVENTION
[0001 ] The present invention concerns information management and retrieval in general.
More specifically, the present invention concerns detecting, and optionally removing, duplicate and near-duplicate information or content, such as in a repository of documents to be searched for example.
1.2 BACKGROUND INFORMATION
[0002] In the following, the term "document(s)" should be broadly interpreted and may include content such as Web pages, text files, multimedia files, object features, link structure, etc. Also, it should be noted that when near-duplicate documents are detected, exact duplicate documents will also be detected as a consequence (though such exact duplicates might not necessarily be distinguished from near-duplicates).
[0003] Detecting near-duplicate documents has many potential applications. For example, duplicate or near-duplicate documents may indicate plagiarism or copyright infringement. One important application of near-duplicate document detection is in the context of information storage and retrieval.
[0004] Efficient techniques to detect documents that are exact duplicates exist. Detecting whether or not documents are near-duplicates is more difficult, particularly in large collections of documents. For example, the Internet, collectively, includes literally billions of "Web 'site"
documents.
documents.
[0005] Sources of duplicate and near-duplicate documents on the Internet are introduced in 1.2.1 below. Then, problems that these duplicate and near-duplicate documents raise, both for end-users and for entities assisting end-users, are introduced in 1.2.2 below. Finally, previous techniques for detecting duplicate and near-duplicate documents in the context of large document collections, as well as perceived shortcomings of such techniques, are introduced in 12.3 below.
1.2.1 SOURCES OF DUPLICATE AND NEAR DUPLICATE
DOCUMENTS ON THE INTERNET
1.2.1 SOURCES OF DUPLICATE AND NEAR DUPLICATE
DOCUMENTS ON THE INTERNET
[0006] On the Internet, the World Wide Web (referred to as "the Web") may include the same document duplicated in different forms or at different places.
(Naturally, other networks, or even stand alone systems, may have duplicate documents.) Sources of such duplication are introduced here.
(Naturally, other networks, or even stand alone systems, may have duplicate documents.) Sources of such duplication are introduced here.
[0007] First, some documents are "mirrored" at different sites on the Web.
Such mirroring is used to alleviate potential delays when many users attempt to request the same document at the same time, and/or to minimize network latency (e.g., by caching Web pages locally).
Such mirroring is used to alleviate potential delays when many users attempt to request the same document at the same time, and/or to minimize network latency (e.g., by caching Web pages locally).
[0008] Second, some documents will have different versions with different formatting. For example, a given document may have plain text and HTML (hyper-text markup language) versions so that users can render or download the content in a form that they prefer. As more and more different devices (e.g., computers, mobile phones, personal digital assistants, etc.) are used to access the Internet, a given document may have more and more different versions with different formatting (text only, text plus other media, etc.).
[0009] Third, documents are often prepended or appended with information related to its location on the Web, the date, the date it was last modified, a version, a title, a hierarchical classification path (e.g., a Web page may be classified under more than one class within the hierarchy of a Web site), etc.
[0010] Fourth, in some instances a new document is generated from an existing document using a consistent word replacement. For example, a Web site may be "re-branded" for different audiences by using word replacement.
[00111 Finally, some Web pages aggregate or incorporate content available from another source on the Web.
1.2.2 PROBLEMS RAISED BY DUPLICATE AND
NEAR-DUPLICATE DOCUMENTS
[00121 Duplicate and near-duplicate documents raise potential problems for both people accessing information (e.g., from the Web) and entities helping people to access desired information (e.g., search engine companies). These potential problems are introduced below.
[0013] Although people continue to use computers to enter, manipulate and store information, in view of developments in data storage, intemetworking (e.g., the Internet), and interlinking and cross referencing of information (e.g., using hyper-text links), people are using computers (or more generally, information access machines) to access information to an ever increasing extent.
10014] Search engines have been employed to help users find desired information. Search engines typically search databased content or "Websites" or "Web pages"
pursuant to a user query. In response to a user's query, a rank-ordered list, which typically includes brief descriptions of the uncovered content, as well as hyper-texts links (i.e., text, having associated URLs) to the uncovered content, is returned. The rank-ordering of the list is typically based on a match between words appearing in the query and words appearing in the content.
[0015] From the perspective of users, duplicate and near-duplicate documents raise problems.
More specifically, when users submit a query to a search engine, most do not want links to (and descriptions of) Web pages which have largely redundant information. For example, search engines typically respond to search queries by providing groups of ten results. If pages with duplicate content were returned, many of the results in one group may include the same content.
Thus, there is a need for techniques to avoid providing search results associated with (e.g., having links to) Web pages having duplicate content.
[0016] From the perspective of entities hosting search engines, duplicate and near-duplicate documents also raise problems -- giving end-users what they want, being one of them. To appreciate some of the other potential problems raised by duplicate and near-duplicate documents, some search engine technology is introduced first.
10017] Most search engines perform three main functions: (i) crawling the Web;
(ii) indexing the content of the Web; and (iii) responding to a search query using the index to generate search results. Given the large amount of information available, these three main functions are automated to a large extent. While the crawl operation will associate words or phrases with a document (e.g., a Web page), the indexing operation will associate document(s) (e.g., Web page(s)) with words or phrases. The search operation then (i) uses that index to find documents (e.g., Web pages) containing various words of a search query, and (ii) ranks or orders the documents found in accordance with some heuristic(s).
10018] Recall that the Web may include the same documents duplicated indifferent forms or at different places on the Web. For example, as introduced in 1.2.1 above, documents may be "mirrored" at different sites on the Web, documents may have a number of different formats so that users can render or download the content in a form that they prefer, documents may have a different versions with different information prepended or appended, some documents may have been generated from others using consistent word replacement, and some documents may aggregate or incorporate documents available from another source on the Web.
It would be desirable to eliminate such duplicates or near-duplicates. Besides eliminating duplicate or near-duplicate documents to meet user expectations and wishes, eliminating duplicate or near-duplicate documents is desirable to search engine hosting entities to (i) reduce storage requirements (e.g., for the index and data structures derived from the index), and (ii) reduce the time and/or computational resources needed to process indexes, queries, etc.
[0019] In view of the foregoing, techniques to detect (and eliminate) near-duplicate documents are needed.
1.2.3 KNOWN TECHNIQUES FOR DETECTING
DUPLICATE AND NEAR-DUPLICATE DOCUMENTS AND
THEIR PERCEIVED LIMITATIONS
10020] A naive solution would be to compare all pairs to documents. Since this is prohibitively expensive on large datasets, Manber (U. Manber, "Finding similar files in a large file system," Proc. of the USENIX Winter 1994 Technical.Conference (Jan.
1994)) and Heintze (N. Heintze, "Scalable Document Fingerprinting," Proc. of the 2nd USENIX
Workshop on Electronic Commerce (Nov 1996)) proposed algorithms for detecting near-duplicate documents that reduced the number of comparisons. Both algorithms work on sequences of adjacent characters. Brin et al. (S. Brin, J. Davis, and H. Garcia-Molina, "Copy Detection Mechanisms for Digital Documents," 1995 ACM SIGMOD International Conference on Management of Data, pp. 398-409 (May 1995)) started to use word sequences to detect copyright violations.
Broder et al. (A. Broder, S. Glassman, M. Manasse, and G. Zweig, "Syntactic Clustering of the Web," 6th International World Wide Web Conference pp. 93-404 (Apr. 1997), also used word sequences to efficiently find near-duplicate Web pages.
Later, Charikar (M. S. Charikar, "Similarity Estimation Techniques from Rounding Algorithms", 34th Annual ACM Symposium on Theory of Computing (May 2002).
See also U.S. Patent Application Publication No. US 2006/0101060 developed an approach based on random projections of the words in a document. Recently Hoad and Zobel (T.C. Hoad and J. Zobel, "Methods for identifying versioned and plagiarized documents", Journal of the American Society for Information Science and Technology, 54(3), pp. 203-215 (2003)) developed and compared methods for identifying versioned and plagiarized documents. Unfortunately, however, the technique recommended by Hoad and Zobel is inefficient, having O(N2) computational complexity, where N is the number of documents to be compared with a document of interest.
1.2.3.1 INTRODUCTION OF THE BRODER AND
CHARIKAR ALGORITHMS FOR DOCUMENT
SIMILARITY
[0021] In both the Broder and Charikar algorithms, each HTML page is converted into a token sequence. The two algorithms differ only in how they convert the token sequence into a bit string representing the page.
[0022] To convert an HTML page into a token sequence, all HTML markup in the page is replaced by white space or, in case of formatting instructions, ignored. Then every maximal alphanumeric sequence is considered a term and is hashed using Rabin's fingerprinting scheme (M. Rabin, "Fingerprinting by random polynomials," Report M-15-8 1, Center for Research in Computing Technology, Harvard University (1981)), to generate tokens, with two exceptions.
[0023] Both algorithms generate a bit string from the token sequence of a page and use it to determine the near-duplicates for the page.
[0024] Let n be the length of the token sequence of a page. Using the Broder algorithm every sub-sequence of k tokens (where the sub-sequences overlap) is fingerprinted using 64-bit Rabin fingerprints, which results in a sequence of (n - k + 1) fingerprints, called "shingles". Let S (d) be the set of shingles of the page "d". The Broder algorithm makes the assumption that the percentage of unique shingles on which the two pages d and d' agree. That is, the Broder JS(d) nS(d')I
algorithm assumes that IS(d)US(d')J is a good measure for the similarity of d and d'.
[0025] The foregoing may be approximated by fingerprinting every shingle with m different fingerprinting functions f; for 1 <_ i 5 m. This leads to (n - k + 1) values for each f;. For each i, the smallest of these values is called "the i-th minvalue" and is stored at the page. As a result, the Broder algorithm creates an m-dimensional vector of minvalues. Broder et al. showed that the expected percentage of entries in the minvalues vector that two pages d and d' agree on is equal to the percentage of unique shingles on which d and d' agree. Thus, to estimate the similarity of two pages, it suffices to determine the percentage of agreeing entries in the minvalues vectors. To save space and speed up the similarity computation, the m-dimensional vector of minvalues might be reduced to an m'-dimensional vector of supershingles by fingerprinting non-overlapping sequences of minvalues. Let m be divisible by m' and let f =
m/m'. The concatenation of minvalue j * ,..., (j + 1) * .e - 1 might be fingerprinted for 0 5 = j < m' with yet another fingerprinting function. (Note that the notion of "megashingles" was also introduced in the Broder et al. paper in order to further speed up the algorithm. Megashingles are generated by fingerprinting every pair of supershingles, such that each rnegashingle is a fingerprinted pair of supershingles. Since, however, using megashingles does not improve precision or recall, they need not be used.) Two pages are near-duplicates under the Broder algorithm (referred to here as "B-similar") if and only if their supershingle vectors agree in at least two supershingles. The number of identical entries in the supershingle vectors of two pages is their "B-similarity". The parameters for the Broder algorithm are m, . , in', and k.
[0026] As can be appreciated from the foregoing, the Broder algorithm is order dependent (e.g., since the shingles are fingerprints of overlapping sub-sequences), but is independent of the frequency of the shingles.
[0027] The Charikar algorithm is now described. Let b be a constant. Each token is projected into b-dimensional space by randomly choosing b entries from {-1,1 }. The resulting b-dimensional vector may be referred to as a "token vector". The same tokens, whether occurring on the same page or on different pages, will have the same b-dimensional representation (i.e., the same "token vector"). For each page, a representative b-dimensional vector (which may be referred to as an "initial page vector") is created by adding the projections of all the tokens in the page's token sequence (i.e., adding all of the page's "token vectors").
The final vector for the page (which may be referred to as a "final page vector") is created by setting every positive entry in the vector to 1 and every non-positive entry to 0. The generates a random b-dimensional projection (i.e., a final page vector) for each page. The final page vectors have the property that the cosine similarity of two pages is proportional to the number of bits in which the two corresponding projections agree. That is, similarity in the Charikar algorithm (referred to as "C-similarity") of two Web pages is the number of bits their projections agree on.
Two pages are near-duplicates in the Charikar algorithm (or are C-similar) if the number of agreeing bits in their projections is above a fixed threshold t.
[0028] As can be appreciated from the foregoing, given the definition of minvalues (from which supershingles; and perhaps even megashingles, are generated), Broder's technique uses representations based on a subset of the words (or tokens) of the original document being analyzed. On the other hand, Charikar's technique uses representations based on all words (not removed by preprocessing) (or tokens) of the original document being analyzed.
That is, Charikar's technique might consider all words (or tokens) of documents accepted as inputs.
Further, Broder's technique uses set intersection to determine whether or not documents are near-duplicates. On the other had, Charikar's technique uses random projections to determine whether or not documents are near-duplicates.
[0029] As can be appreciated from the foregoing, in both algorithms pages with the same token sequence are assigned the same bit string. The Charikar algorithm ignores the order of the tokens (given the additive aspect of generating a page vector from token vectors). The shingles of the Broder algorithm are based on the order of the tokens. However, the Broder algorithm ignores the frequency of shingles. On the other hand, the Charikar algorithm accounts for the frequency of terms (again, given the additive aspect of generating a page vector from token vectors). For both algorithms there can be false positives (non near-duplicate pairs returned as near-duplicates) as well as false negatives (near-duplicate pairs not returned as near-duplicates.) 10030] Let T be the sum of the number of tokens in all documents and let D be the number of documents. The Broder algorithm takes time O(Tm + Dm') = O(Tm). The Broder algorithm takes time O(Tb) to determine the vector for each page. As described below, the C-similar pairs might be computed using a trick similar to supershingles. It takes time O(D) so that the total time for the Charikar algorithm is O(Tb).
100311 Some embodiments consistent with the present invention might further (d) process the set of documents to determine a third set of near-duplicate documents using the second document similarity technique, (e) determine a fourth set of near duplicate documents by determining the union of the second set of near duplicate document and the third set of near-duplicate documents.
1.2.3.2 EVALUATION OF THE BRODER AND
CHARIKAR ALGORITHMS
[0032] The present inventor evaluated the Broder and Charikar algorithms on 1.6B distinct Web pages, according to three criteria -- (1) precision on a random subset, (2) the distribution of the number of term differences per near-duplicate pair, and (3) the distribution of the number of near-duplicates per page. All parameters in the Broder algorithm were set as suggested in the literature. The parameters in the Charikar algorithm were chosen so that it used the same amount of space per document and returned about the same number of correct near-duplicate pairs (i.e., had about the same recall).
[0033] The present inventor found that the Charikar algorithm achieved a precision of 0.50, while the Broder algorithm achieved a precision of 0.38. Both algorithms were found to perform about the same for near-duplicate pairs on the same site (low precision) and for near-duplicate pairs on different sites (high precision). However, over 90% of the near-duplicate pairs found by the Broder algorithm belonged to the same site, but only 74% of the near-duplicate pairs found by the Charikar algorithm belonged to the same site.
(0034] Thus, the Charikar algorithm found more of the near-duplicate pairs for which precision is high. The number of term differences per near-duplicate pair was found to be very similar for the two algorithms, but the Broder algorithm returned fewer pairs with extremely large term differences. The distribution of the number of near-duplicates per page was found to follow a power-law for both algorithms. However, the Broder algorithm was found to have a higher "spread" around the power-law curve. The present inventor believes that a possible reason for that "noise" is that the bit string representing a page in the Broder algorithm is based on a randomly selected subset of terms in the page. Thus, there might be "lucky" and "unlucky"
choices, leading to false near-duplicate pairs or missing actual near-duplicate pairs. The Charikar algorithm does not select a subset of terms but is based on all terms in the page.
[0035] The present inventor found that neither of the algorithms worked well for finding near-duplicate pairs on the same Website, though both achieved high precision for near-duplicate pairs on different Websites.
[0036] In view of the foregoing, it would be useful to provide improved techniques for finding near-duplicate documents. It would be useful if such techniques improved the precision of the Broder and Charikar algorithms. Finally, it would be useful if such techniques worked well for finding near-duplicate pairs on the same Website, as well as on different Websites.
2. SUMMARY OF THE INVENTION
[0037] Some embodiments consistent with the present invention might detect near-duplicate documents by (a) accepting a set of documents, (b) processing the set of documents to determine a first set of near-duplicate documents using a first document similarity technique, and (c) processing the first set of near duplicate documents to determine a second set of near-duplicate documents using 'a second document similarity technique.
[0038] In at least some embodiments consistent with the present invention, the first document similarity technique might be token order dependent, and the second document similarity technique might be order independent.
[0039] In at least some embodiments consistent with the present invention, the first document similarity technique might be token frequency independent, and the second document similarity technique might be frequency dependent.
[0040] In at least some embodiments consistent with the present invention, the first document similarity technique might determine whether two documents are near-duplicates using representations based on a subset of the words or tokens of the documents, and the second document similarity technique might determine whether two documents are near-duplicates using representations based on all of the words or tokens of the documents.
(00411 In at least some embodiments consistent with the present invention, the first document similarity technique might use set intersection to determine whether or not documents are near-duplicates, and the second document similarity technique might use random projections to determine whether or not documents are near-duplicates.
[0042] At least some other embodiments consistent with the present invention might identify near-duplicate documents by (a) accepting a set of documents, and (b) processing the set of documents to determine near-duplicate documents, wherein a first document similarity technique is used to determine near-duplicate documents for documents from the same Website, and wherein a second document similarity technique is used to determine near-duplicate documents for documents from different Websites.
3. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 is a block diagram of an environment in which at least some aspects of the present invention may be used.
[0044] Figure 2 is a process bubble diagram of an advanced search facility in which at least some aspects of the present invention may be used.
[0045] Figure 3 is a flow chart of an exemplary method for determining near duplicate documents in a manner consistent with the present invention.
[0046] Figure 4 is a flow chart of an exemplary method for determining a final set of near duplicate documents from an initial set of near duplicate documents in a manner consistent with the present invention.
[0047] Figure 5 is block diagram of a machine that may be used to perform one or more of the operations discussed above, and/or to store information generated and/or used by such operations, in a manner consistent with the present invention.
4. DETAILED DESCRIPTION
[0048] The present invention may involve novel methods, apparatus, message formats, and/or data structures for determining whether or not documents are similar.
The following description is presented to enable one skilled in the art to make and use the invention, and is provided in the context of particular applications and their requirements.
Thus, the following description of embodiments consistent with the present invention provides illustration and description, but is not intended to be exhaustive or to limit the present invention to the precise form disclosed. Various modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles set forth below may be applied to other embodiments and applications. For example, although a series of acts may be described with reference to a flow diagram, the order of acts may differ in other implementations when the performance of one act is not dependent on the completion of another act.
Further, non-dependent acts may be performed in parallel. No element, act or instruction used in the description should be construed as critical or essential to the present invention unless explicitly described as such. Also, as used herein, the article "a" is intended to include one or more items.
Where only one item is intended, the term "one" or similar language is used.
In the following, "information" may refer to the actual information, or a pointer to, or a location of, such information. Thus, the present invention is not intended to be limited to the embodiments shown and the inventor regards her invention to include any patentable subject matter described.
[00491 In the following, environments in which the present invention may be employed are introduced in 4.1. Then, exemplary embodiments consistent with the present invention are described in 4.2. Finally, some conclusions about the present invention are set forth in 4.3.
4.1 EXEMPLARY ENVIRONMENTS IN WHICH INVENTION
MAY OPERATE
[0050] The following exemplary embodiments are presented to illustrate examples of utility of embodiments consistent with the present invention and to illustrate examples of contexts in which such embodiments may operate. However, the present invention can be used in other environments and its use is not intended to be limited to the exemplary environment 100 and search facility 200 introduced below with reference to Figures 1 and 2, respectively.
[0051] Figure 1 is a block diagram of an environment 100 in which at least some aspects of the present invention may be used. This environment 100 may be a network (such as the Internet for example) 160 in which an information access facility (client) 110 is used to render information accessed from one or more content providers (servers) 180. A
search facility (server) 130 may be used by the information access facility 110 to search for content of interest.
[0052] The information access facility 110 may include a browsing operation 112 which may include a navigation operation 114 and a user interface operation 116. The browsing operation 112 may access the network 160 via input/output interface operations 118. For example, in the context of a personal computer, the browsing operation 112 may be a browser (such as "Firefox"
from Mozilla, "Internet Explorer" from Microsoft Corporation of Redmond, Washington, "Opera" from Opera Software, "Netscape" from Time Warner, Inc.) and the input/output interface operations may include a modem or network interface card (or NIC) and networking software. Other examples of possible information access facilities 110 include untethered devices, such as personal digital assistants and mobile telephones for example, set-top boxes, kiosks, etc.
[0053] Each of the content providers 180 may include stored resources (also referred to as content) 136, a resource retrieval operation 184 that accesses and provides content in response to a request, and input/output interface operation(s) 182. These operations of the content providers 180 may be performed by computers, such as personal computers or servers for example-
[00111 Finally, some Web pages aggregate or incorporate content available from another source on the Web.
1.2.2 PROBLEMS RAISED BY DUPLICATE AND
NEAR-DUPLICATE DOCUMENTS
[00121 Duplicate and near-duplicate documents raise potential problems for both people accessing information (e.g., from the Web) and entities helping people to access desired information (e.g., search engine companies). These potential problems are introduced below.
[0013] Although people continue to use computers to enter, manipulate and store information, in view of developments in data storage, intemetworking (e.g., the Internet), and interlinking and cross referencing of information (e.g., using hyper-text links), people are using computers (or more generally, information access machines) to access information to an ever increasing extent.
10014] Search engines have been employed to help users find desired information. Search engines typically search databased content or "Websites" or "Web pages"
pursuant to a user query. In response to a user's query, a rank-ordered list, which typically includes brief descriptions of the uncovered content, as well as hyper-texts links (i.e., text, having associated URLs) to the uncovered content, is returned. The rank-ordering of the list is typically based on a match between words appearing in the query and words appearing in the content.
[0015] From the perspective of users, duplicate and near-duplicate documents raise problems.
More specifically, when users submit a query to a search engine, most do not want links to (and descriptions of) Web pages which have largely redundant information. For example, search engines typically respond to search queries by providing groups of ten results. If pages with duplicate content were returned, many of the results in one group may include the same content.
Thus, there is a need for techniques to avoid providing search results associated with (e.g., having links to) Web pages having duplicate content.
[0016] From the perspective of entities hosting search engines, duplicate and near-duplicate documents also raise problems -- giving end-users what they want, being one of them. To appreciate some of the other potential problems raised by duplicate and near-duplicate documents, some search engine technology is introduced first.
10017] Most search engines perform three main functions: (i) crawling the Web;
(ii) indexing the content of the Web; and (iii) responding to a search query using the index to generate search results. Given the large amount of information available, these three main functions are automated to a large extent. While the crawl operation will associate words or phrases with a document (e.g., a Web page), the indexing operation will associate document(s) (e.g., Web page(s)) with words or phrases. The search operation then (i) uses that index to find documents (e.g., Web pages) containing various words of a search query, and (ii) ranks or orders the documents found in accordance with some heuristic(s).
10018] Recall that the Web may include the same documents duplicated indifferent forms or at different places on the Web. For example, as introduced in 1.2.1 above, documents may be "mirrored" at different sites on the Web, documents may have a number of different formats so that users can render or download the content in a form that they prefer, documents may have a different versions with different information prepended or appended, some documents may have been generated from others using consistent word replacement, and some documents may aggregate or incorporate documents available from another source on the Web.
It would be desirable to eliminate such duplicates or near-duplicates. Besides eliminating duplicate or near-duplicate documents to meet user expectations and wishes, eliminating duplicate or near-duplicate documents is desirable to search engine hosting entities to (i) reduce storage requirements (e.g., for the index and data structures derived from the index), and (ii) reduce the time and/or computational resources needed to process indexes, queries, etc.
[0019] In view of the foregoing, techniques to detect (and eliminate) near-duplicate documents are needed.
1.2.3 KNOWN TECHNIQUES FOR DETECTING
DUPLICATE AND NEAR-DUPLICATE DOCUMENTS AND
THEIR PERCEIVED LIMITATIONS
10020] A naive solution would be to compare all pairs to documents. Since this is prohibitively expensive on large datasets, Manber (U. Manber, "Finding similar files in a large file system," Proc. of the USENIX Winter 1994 Technical.Conference (Jan.
1994)) and Heintze (N. Heintze, "Scalable Document Fingerprinting," Proc. of the 2nd USENIX
Workshop on Electronic Commerce (Nov 1996)) proposed algorithms for detecting near-duplicate documents that reduced the number of comparisons. Both algorithms work on sequences of adjacent characters. Brin et al. (S. Brin, J. Davis, and H. Garcia-Molina, "Copy Detection Mechanisms for Digital Documents," 1995 ACM SIGMOD International Conference on Management of Data, pp. 398-409 (May 1995)) started to use word sequences to detect copyright violations.
Broder et al. (A. Broder, S. Glassman, M. Manasse, and G. Zweig, "Syntactic Clustering of the Web," 6th International World Wide Web Conference pp. 93-404 (Apr. 1997), also used word sequences to efficiently find near-duplicate Web pages.
Later, Charikar (M. S. Charikar, "Similarity Estimation Techniques from Rounding Algorithms", 34th Annual ACM Symposium on Theory of Computing (May 2002).
See also U.S. Patent Application Publication No. US 2006/0101060 developed an approach based on random projections of the words in a document. Recently Hoad and Zobel (T.C. Hoad and J. Zobel, "Methods for identifying versioned and plagiarized documents", Journal of the American Society for Information Science and Technology, 54(3), pp. 203-215 (2003)) developed and compared methods for identifying versioned and plagiarized documents. Unfortunately, however, the technique recommended by Hoad and Zobel is inefficient, having O(N2) computational complexity, where N is the number of documents to be compared with a document of interest.
1.2.3.1 INTRODUCTION OF THE BRODER AND
CHARIKAR ALGORITHMS FOR DOCUMENT
SIMILARITY
[0021] In both the Broder and Charikar algorithms, each HTML page is converted into a token sequence. The two algorithms differ only in how they convert the token sequence into a bit string representing the page.
[0022] To convert an HTML page into a token sequence, all HTML markup in the page is replaced by white space or, in case of formatting instructions, ignored. Then every maximal alphanumeric sequence is considered a term and is hashed using Rabin's fingerprinting scheme (M. Rabin, "Fingerprinting by random polynomials," Report M-15-8 1, Center for Research in Computing Technology, Harvard University (1981)), to generate tokens, with two exceptions.
[0023] Both algorithms generate a bit string from the token sequence of a page and use it to determine the near-duplicates for the page.
[0024] Let n be the length of the token sequence of a page. Using the Broder algorithm every sub-sequence of k tokens (where the sub-sequences overlap) is fingerprinted using 64-bit Rabin fingerprints, which results in a sequence of (n - k + 1) fingerprints, called "shingles". Let S (d) be the set of shingles of the page "d". The Broder algorithm makes the assumption that the percentage of unique shingles on which the two pages d and d' agree. That is, the Broder JS(d) nS(d')I
algorithm assumes that IS(d)US(d')J is a good measure for the similarity of d and d'.
[0025] The foregoing may be approximated by fingerprinting every shingle with m different fingerprinting functions f; for 1 <_ i 5 m. This leads to (n - k + 1) values for each f;. For each i, the smallest of these values is called "the i-th minvalue" and is stored at the page. As a result, the Broder algorithm creates an m-dimensional vector of minvalues. Broder et al. showed that the expected percentage of entries in the minvalues vector that two pages d and d' agree on is equal to the percentage of unique shingles on which d and d' agree. Thus, to estimate the similarity of two pages, it suffices to determine the percentage of agreeing entries in the minvalues vectors. To save space and speed up the similarity computation, the m-dimensional vector of minvalues might be reduced to an m'-dimensional vector of supershingles by fingerprinting non-overlapping sequences of minvalues. Let m be divisible by m' and let f =
m/m'. The concatenation of minvalue j * ,..., (j + 1) * .e - 1 might be fingerprinted for 0 5 = j < m' with yet another fingerprinting function. (Note that the notion of "megashingles" was also introduced in the Broder et al. paper in order to further speed up the algorithm. Megashingles are generated by fingerprinting every pair of supershingles, such that each rnegashingle is a fingerprinted pair of supershingles. Since, however, using megashingles does not improve precision or recall, they need not be used.) Two pages are near-duplicates under the Broder algorithm (referred to here as "B-similar") if and only if their supershingle vectors agree in at least two supershingles. The number of identical entries in the supershingle vectors of two pages is their "B-similarity". The parameters for the Broder algorithm are m, . , in', and k.
[0026] As can be appreciated from the foregoing, the Broder algorithm is order dependent (e.g., since the shingles are fingerprints of overlapping sub-sequences), but is independent of the frequency of the shingles.
[0027] The Charikar algorithm is now described. Let b be a constant. Each token is projected into b-dimensional space by randomly choosing b entries from {-1,1 }. The resulting b-dimensional vector may be referred to as a "token vector". The same tokens, whether occurring on the same page or on different pages, will have the same b-dimensional representation (i.e., the same "token vector"). For each page, a representative b-dimensional vector (which may be referred to as an "initial page vector") is created by adding the projections of all the tokens in the page's token sequence (i.e., adding all of the page's "token vectors").
The final vector for the page (which may be referred to as a "final page vector") is created by setting every positive entry in the vector to 1 and every non-positive entry to 0. The generates a random b-dimensional projection (i.e., a final page vector) for each page. The final page vectors have the property that the cosine similarity of two pages is proportional to the number of bits in which the two corresponding projections agree. That is, similarity in the Charikar algorithm (referred to as "C-similarity") of two Web pages is the number of bits their projections agree on.
Two pages are near-duplicates in the Charikar algorithm (or are C-similar) if the number of agreeing bits in their projections is above a fixed threshold t.
[0028] As can be appreciated from the foregoing, given the definition of minvalues (from which supershingles; and perhaps even megashingles, are generated), Broder's technique uses representations based on a subset of the words (or tokens) of the original document being analyzed. On the other hand, Charikar's technique uses representations based on all words (not removed by preprocessing) (or tokens) of the original document being analyzed.
That is, Charikar's technique might consider all words (or tokens) of documents accepted as inputs.
Further, Broder's technique uses set intersection to determine whether or not documents are near-duplicates. On the other had, Charikar's technique uses random projections to determine whether or not documents are near-duplicates.
[0029] As can be appreciated from the foregoing, in both algorithms pages with the same token sequence are assigned the same bit string. The Charikar algorithm ignores the order of the tokens (given the additive aspect of generating a page vector from token vectors). The shingles of the Broder algorithm are based on the order of the tokens. However, the Broder algorithm ignores the frequency of shingles. On the other hand, the Charikar algorithm accounts for the frequency of terms (again, given the additive aspect of generating a page vector from token vectors). For both algorithms there can be false positives (non near-duplicate pairs returned as near-duplicates) as well as false negatives (near-duplicate pairs not returned as near-duplicates.) 10030] Let T be the sum of the number of tokens in all documents and let D be the number of documents. The Broder algorithm takes time O(Tm + Dm') = O(Tm). The Broder algorithm takes time O(Tb) to determine the vector for each page. As described below, the C-similar pairs might be computed using a trick similar to supershingles. It takes time O(D) so that the total time for the Charikar algorithm is O(Tb).
100311 Some embodiments consistent with the present invention might further (d) process the set of documents to determine a third set of near-duplicate documents using the second document similarity technique, (e) determine a fourth set of near duplicate documents by determining the union of the second set of near duplicate document and the third set of near-duplicate documents.
1.2.3.2 EVALUATION OF THE BRODER AND
CHARIKAR ALGORITHMS
[0032] The present inventor evaluated the Broder and Charikar algorithms on 1.6B distinct Web pages, according to three criteria -- (1) precision on a random subset, (2) the distribution of the number of term differences per near-duplicate pair, and (3) the distribution of the number of near-duplicates per page. All parameters in the Broder algorithm were set as suggested in the literature. The parameters in the Charikar algorithm were chosen so that it used the same amount of space per document and returned about the same number of correct near-duplicate pairs (i.e., had about the same recall).
[0033] The present inventor found that the Charikar algorithm achieved a precision of 0.50, while the Broder algorithm achieved a precision of 0.38. Both algorithms were found to perform about the same for near-duplicate pairs on the same site (low precision) and for near-duplicate pairs on different sites (high precision). However, over 90% of the near-duplicate pairs found by the Broder algorithm belonged to the same site, but only 74% of the near-duplicate pairs found by the Charikar algorithm belonged to the same site.
(0034] Thus, the Charikar algorithm found more of the near-duplicate pairs for which precision is high. The number of term differences per near-duplicate pair was found to be very similar for the two algorithms, but the Broder algorithm returned fewer pairs with extremely large term differences. The distribution of the number of near-duplicates per page was found to follow a power-law for both algorithms. However, the Broder algorithm was found to have a higher "spread" around the power-law curve. The present inventor believes that a possible reason for that "noise" is that the bit string representing a page in the Broder algorithm is based on a randomly selected subset of terms in the page. Thus, there might be "lucky" and "unlucky"
choices, leading to false near-duplicate pairs or missing actual near-duplicate pairs. The Charikar algorithm does not select a subset of terms but is based on all terms in the page.
[0035] The present inventor found that neither of the algorithms worked well for finding near-duplicate pairs on the same Website, though both achieved high precision for near-duplicate pairs on different Websites.
[0036] In view of the foregoing, it would be useful to provide improved techniques for finding near-duplicate documents. It would be useful if such techniques improved the precision of the Broder and Charikar algorithms. Finally, it would be useful if such techniques worked well for finding near-duplicate pairs on the same Website, as well as on different Websites.
2. SUMMARY OF THE INVENTION
[0037] Some embodiments consistent with the present invention might detect near-duplicate documents by (a) accepting a set of documents, (b) processing the set of documents to determine a first set of near-duplicate documents using a first document similarity technique, and (c) processing the first set of near duplicate documents to determine a second set of near-duplicate documents using 'a second document similarity technique.
[0038] In at least some embodiments consistent with the present invention, the first document similarity technique might be token order dependent, and the second document similarity technique might be order independent.
[0039] In at least some embodiments consistent with the present invention, the first document similarity technique might be token frequency independent, and the second document similarity technique might be frequency dependent.
[0040] In at least some embodiments consistent with the present invention, the first document similarity technique might determine whether two documents are near-duplicates using representations based on a subset of the words or tokens of the documents, and the second document similarity technique might determine whether two documents are near-duplicates using representations based on all of the words or tokens of the documents.
(00411 In at least some embodiments consistent with the present invention, the first document similarity technique might use set intersection to determine whether or not documents are near-duplicates, and the second document similarity technique might use random projections to determine whether or not documents are near-duplicates.
[0042] At least some other embodiments consistent with the present invention might identify near-duplicate documents by (a) accepting a set of documents, and (b) processing the set of documents to determine near-duplicate documents, wherein a first document similarity technique is used to determine near-duplicate documents for documents from the same Website, and wherein a second document similarity technique is used to determine near-duplicate documents for documents from different Websites.
3. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 is a block diagram of an environment in which at least some aspects of the present invention may be used.
[0044] Figure 2 is a process bubble diagram of an advanced search facility in which at least some aspects of the present invention may be used.
[0045] Figure 3 is a flow chart of an exemplary method for determining near duplicate documents in a manner consistent with the present invention.
[0046] Figure 4 is a flow chart of an exemplary method for determining a final set of near duplicate documents from an initial set of near duplicate documents in a manner consistent with the present invention.
[0047] Figure 5 is block diagram of a machine that may be used to perform one or more of the operations discussed above, and/or to store information generated and/or used by such operations, in a manner consistent with the present invention.
4. DETAILED DESCRIPTION
[0048] The present invention may involve novel methods, apparatus, message formats, and/or data structures for determining whether or not documents are similar.
The following description is presented to enable one skilled in the art to make and use the invention, and is provided in the context of particular applications and their requirements.
Thus, the following description of embodiments consistent with the present invention provides illustration and description, but is not intended to be exhaustive or to limit the present invention to the precise form disclosed. Various modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles set forth below may be applied to other embodiments and applications. For example, although a series of acts may be described with reference to a flow diagram, the order of acts may differ in other implementations when the performance of one act is not dependent on the completion of another act.
Further, non-dependent acts may be performed in parallel. No element, act or instruction used in the description should be construed as critical or essential to the present invention unless explicitly described as such. Also, as used herein, the article "a" is intended to include one or more items.
Where only one item is intended, the term "one" or similar language is used.
In the following, "information" may refer to the actual information, or a pointer to, or a location of, such information. Thus, the present invention is not intended to be limited to the embodiments shown and the inventor regards her invention to include any patentable subject matter described.
[00491 In the following, environments in which the present invention may be employed are introduced in 4.1. Then, exemplary embodiments consistent with the present invention are described in 4.2. Finally, some conclusions about the present invention are set forth in 4.3.
4.1 EXEMPLARY ENVIRONMENTS IN WHICH INVENTION
MAY OPERATE
[0050] The following exemplary embodiments are presented to illustrate examples of utility of embodiments consistent with the present invention and to illustrate examples of contexts in which such embodiments may operate. However, the present invention can be used in other environments and its use is not intended to be limited to the exemplary environment 100 and search facility 200 introduced below with reference to Figures 1 and 2, respectively.
[0051] Figure 1 is a block diagram of an environment 100 in which at least some aspects of the present invention may be used. This environment 100 may be a network (such as the Internet for example) 160 in which an information access facility (client) 110 is used to render information accessed from one or more content providers (servers) 180. A
search facility (server) 130 may be used by the information access facility 110 to search for content of interest.
[0052] The information access facility 110 may include a browsing operation 112 which may include a navigation operation 114 and a user interface operation 116. The browsing operation 112 may access the network 160 via input/output interface operations 118. For example, in the context of a personal computer, the browsing operation 112 may be a browser (such as "Firefox"
from Mozilla, "Internet Explorer" from Microsoft Corporation of Redmond, Washington, "Opera" from Opera Software, "Netscape" from Time Warner, Inc.) and the input/output interface operations may include a modem or network interface card (or NIC) and networking software. Other examples of possible information access facilities 110 include untethered devices, such as personal digital assistants and mobile telephones for example, set-top boxes, kiosks, etc.
[0053] Each of the content providers 180 may include stored resources (also referred to as content) 136, a resource retrieval operation 184 that accesses and provides content in response to a request, and input/output interface operation(s) 182. These operations of the content providers 180 may be performed by computers, such as personal computers or servers for example-
11 Accordingly, the stored resources 186 may be embodied as data stored on some type of storage medium such as a magnetic disk(s), an optical disk(s), etc. In this particular environment 100, the term "document" may be interpreted to include addressable content, such as a Web page for example.
[0054] The search facility 130 may perform crawling, indexing/sorting, and query processing functions. These functions may be performed by the same entity or separate entities. Further, these functions may be performed at the same location or at different locations. In any event, at a crawling facility 150, a crawling operation 152 gets content from various sources accessible via the network 160, and stores such content, or a form of such content, as indicated by 154.
Then, at an automated indexing/sorting facility 140, an automated indexing/sorting operation 142 may access the stored content 154 and may generate a content index (e.g., an inverted index, to be described below) and content ratings (e.g., PageRanks, to be described below) 140.
Finally, a query processing operation 134 accepts queries and returns query results based on the content index (and the content ratings) 140. The crawling, indexing/sorting and query processing functions may be performed by one or more computers.
[0055] Although embodiments consistent with the present invention may be used with a number of different types of search engines, it might be used with an advanced search facility, such as the one presently available from Google, Inc. of Mountain View, California. Figure 2 is a process bubble diagram of such an advanced search facility 200 in which at least some aspects of embodiments consistent with the present invention may be used.
[0056] The advanced search facility 200 illustrated in Figure 2 performs three main functions: (i) crawling; (ii) indexing/sorting; and (iii) searching. The horizontal dashed lines divide Figure 2 into three parts corresponding to these three main functions.
More specifically, the first part 150' corresponds to the crawling function, the second part 140' corresponds to the indexing/sorting function, and the third part 134' corresponds to the search (or query processing) function. (Note that an apostrophe following a reference number is used to indicate that the referenced item is merely one example of the item referenced by the number without an apostrophe.) Each of these parts is introduced in more detail below. Before doing so, however, a few distinguishing features of this advanced search facility 200 are introduced.
[0057] The advanced search facility uses the link structure of the Web, as well as other techniques, to improve search results. (See, e.g., U.S. Patent No. 6,285,999 and the article S.
Brin and L. Page, "The Anatomy of a Large-Scale Hypertextual Search Engine,"
Seventh
[0054] The search facility 130 may perform crawling, indexing/sorting, and query processing functions. These functions may be performed by the same entity or separate entities. Further, these functions may be performed at the same location or at different locations. In any event, at a crawling facility 150, a crawling operation 152 gets content from various sources accessible via the network 160, and stores such content, or a form of such content, as indicated by 154.
Then, at an automated indexing/sorting facility 140, an automated indexing/sorting operation 142 may access the stored content 154 and may generate a content index (e.g., an inverted index, to be described below) and content ratings (e.g., PageRanks, to be described below) 140.
Finally, a query processing operation 134 accepts queries and returns query results based on the content index (and the content ratings) 140. The crawling, indexing/sorting and query processing functions may be performed by one or more computers.
[0055] Although embodiments consistent with the present invention may be used with a number of different types of search engines, it might be used with an advanced search facility, such as the one presently available from Google, Inc. of Mountain View, California. Figure 2 is a process bubble diagram of such an advanced search facility 200 in which at least some aspects of embodiments consistent with the present invention may be used.
[0056] The advanced search facility 200 illustrated in Figure 2 performs three main functions: (i) crawling; (ii) indexing/sorting; and (iii) searching. The horizontal dashed lines divide Figure 2 into three parts corresponding to these three main functions.
More specifically, the first part 150' corresponds to the crawling function, the second part 140' corresponds to the indexing/sorting function, and the third part 134' corresponds to the search (or query processing) function. (Note that an apostrophe following a reference number is used to indicate that the referenced item is merely one example of the item referenced by the number without an apostrophe.) Each of these parts is introduced in more detail below. Before doing so, however, a few distinguishing features of this advanced search facility 200 are introduced.
[0057] The advanced search facility uses the link structure of the Web, as well as other techniques, to improve search results. (See, e.g., U.S. Patent No. 6,285,999 and the article S.
Brin and L. Page, "The Anatomy of a Large-Scale Hypertextual Search Engine,"
Seventh
12 = WO 2008/019133 PCT/US2007/017487 International World Wide Web Conference, Brisbane, Australia.
[0058] Referring back to Figure 2, the three main parts of the advanced search engine 200 are now described further.
[0059] The crawling part 150' may be distributed across a number of machines.
A single URLserver (not shown) serves lists of uniform resource locations ("URLs") 206 to a number of crawlers. Based on this list of URLs 206, the crawling operation 202 crawls the network 160' and gets Web pages 208. A pre-indexing operation 210 may then generate page rankings 212, as well as a repository 214 from these Web pages 208. The page rankings 212 may include a number of URL fingerprint (i.e., a unique value), PageRank value (is introduced above) pairs.
The repository 214 may include URL, content type and compressed page triples.
[0060] Regarding the indexing/sorting part 140', the indexing/sorting operations 220 may generate an inverted index 226. The indexing/sorting operations 220 may also generate page ranks 228 from the citation rankings 212. The page ranks 228 may include document ID, PageRank value pairs.
[0061 ] Regarding the query processing part 134', the searching operations 230 may be run by a Web server and may use a lexicon 232, together with the inverted index 226 and the PageRanks 228, to generate query results in response to a query. The query results may be based on a combination of (i) information derived from PageRanks 228 and (ii) information derived from how closely a particular document matches the terms contained in the query (also referred to as the information retrieval (or "IR") component). Having described exemplary environments in which the present invention may be used, exemplary embodiments consistent with the present invention are now described in 4.2 below.
4.2 EXEMPLARY EMBODIMENTS
[0062] Exemplary methods consistent with the present invention are described in 4.2.1 below. Then, exemplary apparatus consistent with the present invention are described in 4.2.2 below. Finally, refinements alternative and extensions of such embodiments are described in 4.2.3 below.
[0058] Referring back to Figure 2, the three main parts of the advanced search engine 200 are now described further.
[0059] The crawling part 150' may be distributed across a number of machines.
A single URLserver (not shown) serves lists of uniform resource locations ("URLs") 206 to a number of crawlers. Based on this list of URLs 206, the crawling operation 202 crawls the network 160' and gets Web pages 208. A pre-indexing operation 210 may then generate page rankings 212, as well as a repository 214 from these Web pages 208. The page rankings 212 may include a number of URL fingerprint (i.e., a unique value), PageRank value (is introduced above) pairs.
The repository 214 may include URL, content type and compressed page triples.
[0060] Regarding the indexing/sorting part 140', the indexing/sorting operations 220 may generate an inverted index 226. The indexing/sorting operations 220 may also generate page ranks 228 from the citation rankings 212. The page ranks 228 may include document ID, PageRank value pairs.
[0061 ] Regarding the query processing part 134', the searching operations 230 may be run by a Web server and may use a lexicon 232, together with the inverted index 226 and the PageRanks 228, to generate query results in response to a query. The query results may be based on a combination of (i) information derived from PageRanks 228 and (ii) information derived from how closely a particular document matches the terms contained in the query (also referred to as the information retrieval (or "IR") component). Having described exemplary environments in which the present invention may be used, exemplary embodiments consistent with the present invention are now described in 4.2 below.
4.2 EXEMPLARY EMBODIMENTS
[0062] Exemplary methods consistent with the present invention are described in 4.2.1 below. Then, exemplary apparatus consistent with the present invention are described in 4.2.2 below. Finally, refinements alternative and extensions of such embodiments are described in 4.2.3 below.
13 4.2.1 EXEMPLARY METHODS
[0063] Figure 3 is a flow chart of an exemplary method 300 for determining near duplicate documents in a manner consistent with the present invention. As shown, the exemplary method 300 accepts a set of documents. (Block 310) The set of documents might then be processed to determine an initial set of near duplicate documents using a first document similarity technique.
(Block 320). Finally, the initial set of near duplicate documents might be processed to determine a final set of near duplicate documents using a second document similarity technique (Block 330) before the method 300 is left (Node 340).
[0064] Referring back to block 310, in some embodiments consistent with the present invention, the documents might be Web pages. In other embodiments consistent with the present invention, the documents might be a set of token sequence bit strings derived from source documents, such as Web pages for example.
[0065] Referring back to block 320, in some embodiments consistent with the present invention, the first document similarity technique might be order dependent, and/or frequency independent (e.g., with respect to document words, document n-grams, document tokens, etc.).
[0066] Still referring to block 320, in some embodiments consistent with the present invention, the first document similarity technique might include (i) fingerprinting every sub-sequence of k tokens to generate (n - k + 1) shingles, (ii) fingerprinting each shingle with m different fingerprinting functions $ for 1 S i:5 m to generate (n - k + 1) values for each of the m fingerprinting functions f;, (iii) determining, for each i, the smallest value to create an m_ dimensional vector of minvalues, (iv) reducing the m-dimensional vector of minvalues to an m'-dimensional vector of supershingles by fingerprinting non-overlapping sequences of ninvalues, and (v) concluding that two documents are near-duplicates if and only if their supershingle vectors agree in at least two supershingles (or, alternatively, if the two documents agree in at least one megashingle). The first document similarity technique might be any of the techniques described in the Broder paper introduced in 1.2.3 above or described in the Fetterly papers (D. Fetterly, M. Manasse, and M. Najork, "On the Evolution of Clusters of Near-Duplicate Web Pages," 1st Latin American Web Con ess, pp.37-45 (Nov. 2003);
and D.
Fetterly, M. Manasse, and M. Najork, "Detecting Phrase-Level Duplication on the World Wide Web," 28th Annual International ACM SIGIR Conference (Aug. 2005).
In such embodiments, the parameter m might be set to 84, the parameter 1
[0063] Figure 3 is a flow chart of an exemplary method 300 for determining near duplicate documents in a manner consistent with the present invention. As shown, the exemplary method 300 accepts a set of documents. (Block 310) The set of documents might then be processed to determine an initial set of near duplicate documents using a first document similarity technique.
(Block 320). Finally, the initial set of near duplicate documents might be processed to determine a final set of near duplicate documents using a second document similarity technique (Block 330) before the method 300 is left (Node 340).
[0064] Referring back to block 310, in some embodiments consistent with the present invention, the documents might be Web pages. In other embodiments consistent with the present invention, the documents might be a set of token sequence bit strings derived from source documents, such as Web pages for example.
[0065] Referring back to block 320, in some embodiments consistent with the present invention, the first document similarity technique might be order dependent, and/or frequency independent (e.g., with respect to document words, document n-grams, document tokens, etc.).
[0066] Still referring to block 320, in some embodiments consistent with the present invention, the first document similarity technique might include (i) fingerprinting every sub-sequence of k tokens to generate (n - k + 1) shingles, (ii) fingerprinting each shingle with m different fingerprinting functions $ for 1 S i:5 m to generate (n - k + 1) values for each of the m fingerprinting functions f;, (iii) determining, for each i, the smallest value to create an m_ dimensional vector of minvalues, (iv) reducing the m-dimensional vector of minvalues to an m'-dimensional vector of supershingles by fingerprinting non-overlapping sequences of ninvalues, and (v) concluding that two documents are near-duplicates if and only if their supershingle vectors agree in at least two supershingles (or, alternatively, if the two documents agree in at least one megashingle). The first document similarity technique might be any of the techniques described in the Broder paper introduced in 1.2.3 above or described in the Fetterly papers (D. Fetterly, M. Manasse, and M. Najork, "On the Evolution of Clusters of Near-Duplicate Web Pages," 1st Latin American Web Con ess, pp.37-45 (Nov. 2003);
and D.
Fetterly, M. Manasse, and M. Najork, "Detecting Phrase-Level Duplication on the World Wide Web," 28th Annual International ACM SIGIR Conference (Aug. 2005).
In such embodiments, the parameter m might be set to 84, the parameter 1
14 might be set to 14, the parameter m' might be set to 6 and the parameter k might be set to any value from 5 to 10 (e.g., 8). Some embodiments consistent with the present invention might use the following parameter values: m=84; a =14; m'=6; and k= some value from 5 to 10. = Some embodiments consistent with the present invention might omit the wrapping of the shingling "window" from the end of the document, in which case (n - k + 1) shingles are generated. If, on the other hand, the shingling window can wrap around from the end of the document to its beginning, n shingles are generated.
[0067] Referring back to block 330, in some embodiments consistent with the present invention, the second document similarity technique might be order independent, and/or frequency dependent (e.g., with respect to document words, document n-grams, document tokens, etc.).
[0068] Still referring to block 330, in some embodiments consistent with the present invention, the set of documents might be a set of token sequence bit strings, and the second document similarity technique might include (i) projecting each token into b-dimensional space by randomly choosing a predetermined number b of entries from (-1,1), (ii) for each document, creating a b-dimensional vector by adding the projections of all the tokens in its token sequence, and creating a final vector for the document by setting every positive entry in the b-dimensional vector to I and every non-positive entry to 0, and (iii) determining a similarity between two documents based on a number of bits that agree in corresponding projections of the two documents. The second document similarity technique might be any of the techniques described in the Charikar paper introduced in 1.2.3 above. In such embodiments, the parameter b might be set such that a bit string of 48 bytes is stored per document (e.g., b might be set to 384). Naturally, some embodiments consistent with the present invention might select a value for the parameter b to use less (or more) space.
[0069] In some embodiments consistent with the present invention, the second technique might operate on overlapping sequences of tokens (i.e., shingles) instead of on individual tokens.
[0070] Still referring to block 330 of Figure 3, Figure 4 is a flow chart of an exemplary method 400 for determining a final set of near duplicate documents from an initial set of near duplicate documents in a manner consistent with the present invention. As shown, the exemplary method 400 might accept the initial set of near duplicate documents.
(Block 410) [0071 ] As indicated by loop 420-460, a number of acts might be performed for each pair of near duplicate documents in the initial set. Specifically, a similarity value might be determined using the second document similarity technique. (Block 430), Whether the determined similarity value is less than a threshold might then be determined. (Decision block 440) If it is determined that the determined similarity value is less than the threshold, then the current pair of near duplicate documents might be removed from the initial set to generate an updated set (Block 450) before the method 400 continues to block 460. If, on the other hand, it is determined that the determined similarity value is not less than the threshold, then the method 400 might directly proceed to block 460.
[0072] Referring to block 460, once all of the pairs of near duplicate documents in the initial set have been processed, the final set of near duplicate documents might be set to the most recent updated set of near duplicate documents (or to the initial set of near duplicate documents in the event that the determined similarity value was never less than the threshold) (Block 470) before the method 400 is left (Node 480).
[0073] Referring back to block 410, in some embodiments consistent with the present invention, the set of documents might be Web pages. Alternatively, the set of documents might be a set of token sequence bit strings.
[0074] Referring back to block 430, in some embodiments consistent with the present invention, the second document similarity technique might include (i) projecting each of a number of token sequence bit strings into b-dimensional space by randomly choosing- a predetermined number b of entries from {-1, 11, (ii) for each document, creating a b-dimensional vector by adding the projections of all the tokens in its token sequence, and creating a final vector for the document by setting every positive entry in the b-dimensional vector to 1 and every non-positive entry to 0, and (iii) determining a similarity between two documents based on a number of bits that agree in corresponding projections of the two documents.
Referring back to block 440, in such an embodiment, the predetermined number b might be 384 and the threshold might be set to 372. In some embodiments consistent with the present invention, the threshold might be set to approximately 97% (or at least 96%) of the predetermined number b. The predetermined number b might be a lower value. For example, setting b to values as low as 192 has provided good results. Indeed, the present inventor believes that setting b to values of 100 or even slightly less might provide adequate results. In these lower settings of b, the threshold might be set to about 97% of b, or at least 96% of b.
= WO 2008/019133 PCTIUS2007/017487 4.2.2 EXEMPLARY APPARATUS
(0075] Figure 5 is block diagram of a machine 500 that may be used to perform one or more of the operations discussed above, and/or to store information generated and/or used by such operations, in a manner consistent with the present invention. The machine 500 basically includes a processor(s) 510, an input/output interface unit(s) 530, a storage device(s) 520, and a system bus or network 540 for facilitating the communication of information among the coupled elements. An input device(s) 532 and an output device(s) 534 may be coupled with the input/output interface(s) 530.
[00761 The processor(s) 510 may execute machine-executable instructions to affect one or more aspects of the present invention. At least a portion of the machine executable instructions may be stored (temporarily or more permanently) on the storage device(s) 520 and/or may be received from an external source via an input interface unit 530.
[0077] Some aspects of exemplary embodiments consistent with the present invention may be performed in the general context of computer-executable instructions, such as program modules, being executed by a personal computer. However, methods may be performed by (and data structures may be stored on) other apparatus. Program modules may include routines, programs, objects, components, data structures, etc. that perform an operation(s) or implement particular abstract data types. Moreover, those skilled in the art will appreciate that at least some aspects of the present invention may be practiced with other configurations, including hand-held devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, network computers, minicomputers, set-top boxes, mainframe computers, and the like. At least some aspects of the present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices linked through a communications network. In a distributed computing environment, program modules may be located in local and/or remote memory storage devices.
[0078] In one embodiment consistent with the present invention, the machine 500 may be one or more conventional personal computers or servers. In this case, the processing unit(s) 510 may be one or more microprocessors, the bus 540 may include a system bus that couples various system components including a system memory to the processing unit(s). The system bus 540 may be any of several types of bus structures including a memory bus or memory controller, a peripheral bus, and a local bus using any of a variety of bus architectures.
The storage devices 520 may include system memory, such as read only memory (ROM) and/or random access memory (RAM). A basic input/output system (BIOS), containing basic routines that help to transfer information between elements within the personal computer, such as during start-up, may be stored in ROM. The storage device(s) 520 may also include a hard disk drive for reading from and writing to a hard disk, a magnetic disk drive for reading from or writing to a (e.g., removable) magnetic disk, and an optical disk drive for reading from or writing to a removable (magneto-) optical disk such as a compact disk or other (magneto-) optical media.
The hard disk drive, magnetic disk drive, and (magneto-) optical disk drive may be coupled with the system bus 540 by a hard disk drive interface, a magnetic disk drive interface, and an (magneto-) optical drive interface, respectively. The drives and their associated storage media may provide nonvolatile storage of machine-readable instructions, data structures, program modules and other data for the personal computer. Although the exemplary environment described herein employs a hard disk, a removable magnetic disk and a removable optical disk, those skilled in the art will appreciate that other types of storage media (with appropriate interface devices), may be used instead of, or in addition to, the storage devices introduced above.
[0079] A user may enter commands and information into the personal computer through input devices 532, such as a keyboard and pointing device (e.g., a mouse) for example. Other input devices such as a microphone, a joystick, a game pad, a satellite dish, a scanner, or the like, may also (or alternatively) be included. These and other input devices are often connected to the processing unit(s) 510 through a serial port interface 530 coupled to the system bus 540.
Input devices may be connected by other interfaces 530, such as a parallel port, a game port or a universal serial bus (USB). However, in the context of a search facility 130, no input devices, other than those needed to accept queries, and possibly those for system administration and maintenance, are needed.
[0080] The output device(s) 534 may include a monitor or other type of display device, which may also be connected to the system bus 540 via an interface 530, such as a video adapter for example. In addition to (or instead of) the monitor, the personal computer may include other (peripheral) output devices (not shown), such as speakers and printers for example. Again, in the context of a search facility 130, no output devices, other than those needed to communicate query results, and possibly those for system administration and maintenance, are needed.
[0081] The computer may operate in a networked environment which defines logical and/or physical connections to one or more remote computers, such as a remote computer. The remote 1s computer may be another personal computer, a server, a router, a network computer, a peer device or other common network node, and may include many or all of the elements described above relative to the personal computer. The logical and/or physical connections may include a local area network (LAN) and a wide area network (WAN). An intranet and the Internet may be used instead of, or in addition to, such networks.
[0082] When used in a LAN, the personal computer may be connected to the LAN
through a network interface adapter (or "NIC") 530. When used in a WAN, such as the Internet, the personal computer may include a modem or other means for establishing communications over the wide area network. In a networked environment, at least some of the program modules depicted relative to the personal computer may be stored in the remote memory storage device.
The network connections shown are exemplary and other means of establishing a communications link between the computers may be used.
[0083] Referring once again to Figure 1, the information access facility 110 maybe a personal computer, the browsing operation 112 may be an Internet browser, and the input/output interface operation(s) 118 may include communications software and hardware.
Other information access facilities 110 may be untethered devices such as mobile telephones, personal digital assistants, etc., or other information appliances such as set-top boxes, network appliances, etc.
4.2.3 REFINEMENTS, ALTERNATIVES AND EXTENSIONS
[0084] Although embodiments consistent with the present invention were described as processing Web page documents, embodiments consistent with the present invention can operate on various other types of documents (e.g., snippets extracted from other documents, text documents, spreadsheets, database records, media streams, emails, email SPAM, bit sequences, genetic sequences, nucleotide sequences, representations of chemical structures, representations of molecular structures, characteristics of physical objects, etc.). Thus, embodiments consistent with the present invention have various applications. For example, embodiments consistent with the present invention might be used for spam detection, since spam is often replicated many times, perhaps with subtle differences. As another example, embodiments consistent with the present invention might be use to detect redundant snippets of news stories.
[0085] Some near-duplicate document detection algorithms perform poorly on pairs of Web pages from the same Website. The present inventor believes that this is mostly due to =
WO 2008/019133 PCT[US2007/017487 boilerplate text. In some alternative embodiments consistent with the present invention, boilerplate might be detected, and then removed or ignored in near-duplicate document analysis.
Alternatively, or in addition, an algorithm used to analyze Web pages on the same Website to find near-duplicate documents might be different (and potentially slower) than another algorithm used to analyze pairs of Web pages on different Websites.
[0086] Referring back to block 320 of Figure 3, some embodiments consistent with the present invention might modify the first document similarity technique such that features are weighted by frequency.
[0087] Still referring to Figure 3, although the exemplary methods consistent with the present invention described the simple case of a first document similarity technique followed by a second document similarity technique, more than two document similarity techniques might be used. Alternatively, or in addition, at least two document similarity techniques might be used, at least one of which might be applied recursively.
[0088] In some exemplary embodiments described above, a Charikar-based technique was run after a Broder-Fetterly-based technique. This is because a Charikar-based technique can be tuned to a finer degree (e.g., 372 bits of 384 bit vectors match) than a Broder-Fetterly-based technique (e.g., 2 of 6 matching supershingles). Thus, in such embodiments, the second technique can be tuned to a finer degree than the first technique. However, in alternative embodiments consistent with the present invention, other considerations might be used in determining which technique to run first. For example, in some embodiments consistent with the present invention, the second technique might take longer to run (and/or require more storage) than the first technique.
[0089] Recall that documents may be processed to generate tokens. Some embodiments consistent with the present invention might apply special processing to URLs and/or images.
For example, in some embodiments consistent with the present invention, every URL contained in the text of the page might be broken at slashes and dots, and treated like a sequence of individual terms. In some embodiments consistent with the present invention, in order to distinguish pages with different images, the URL in an IMG-tag might be considered to be a term in the page. More specifically, if the URL points to a different host, the whole URL might be considered to be a term. If, on the other hand, it points to the host of the page itself, only the filename of the URL might be used as term. Thus if a page and its images on the same host are mirrored on a different host, the URLs of the IMG-tags might generate the same tokens in the original and mirrored version. URLs can be processed using other, alternative techniques.
=
Indeed, some embodiments consistent with the present invention might ignore URLs, or simply treat URLs as a term.
[0090] Although exemplary embodiments described above might use Rabin's fingerprinting technique to generate tokens, tokens can be generated using other fingerprinting techniques (e.g., fingerprinting techniques referenced in the Hoad and Zobel paper).
[0091] Some embodiments consistent with the present invention might increase recall by, in addition to determining a "final" set of near-duplicate documents as described above, determining a second final set of near-duplicate documents using a Charikar-based technique (preferably with a high threshold, such as 97% of b, or at least 96% of b). A
union of the final set and second final set of near duplicate documents is taken to obtain a "high recall" set of near duplicate documents.
4.3 CONCLUSIONS
[0092] As can be appreciated from the foregoing, improved near-duplicate detection techniques are disclosed. These near-duplicate detection techniques performed well, particularly when analyzing Web pages from the same Website. These techniques did so without sacrificing much in the number of returned correct pairs.
[0067] Referring back to block 330, in some embodiments consistent with the present invention, the second document similarity technique might be order independent, and/or frequency dependent (e.g., with respect to document words, document n-grams, document tokens, etc.).
[0068] Still referring to block 330, in some embodiments consistent with the present invention, the set of documents might be a set of token sequence bit strings, and the second document similarity technique might include (i) projecting each token into b-dimensional space by randomly choosing a predetermined number b of entries from (-1,1), (ii) for each document, creating a b-dimensional vector by adding the projections of all the tokens in its token sequence, and creating a final vector for the document by setting every positive entry in the b-dimensional vector to I and every non-positive entry to 0, and (iii) determining a similarity between two documents based on a number of bits that agree in corresponding projections of the two documents. The second document similarity technique might be any of the techniques described in the Charikar paper introduced in 1.2.3 above. In such embodiments, the parameter b might be set such that a bit string of 48 bytes is stored per document (e.g., b might be set to 384). Naturally, some embodiments consistent with the present invention might select a value for the parameter b to use less (or more) space.
[0069] In some embodiments consistent with the present invention, the second technique might operate on overlapping sequences of tokens (i.e., shingles) instead of on individual tokens.
[0070] Still referring to block 330 of Figure 3, Figure 4 is a flow chart of an exemplary method 400 for determining a final set of near duplicate documents from an initial set of near duplicate documents in a manner consistent with the present invention. As shown, the exemplary method 400 might accept the initial set of near duplicate documents.
(Block 410) [0071 ] As indicated by loop 420-460, a number of acts might be performed for each pair of near duplicate documents in the initial set. Specifically, a similarity value might be determined using the second document similarity technique. (Block 430), Whether the determined similarity value is less than a threshold might then be determined. (Decision block 440) If it is determined that the determined similarity value is less than the threshold, then the current pair of near duplicate documents might be removed from the initial set to generate an updated set (Block 450) before the method 400 continues to block 460. If, on the other hand, it is determined that the determined similarity value is not less than the threshold, then the method 400 might directly proceed to block 460.
[0072] Referring to block 460, once all of the pairs of near duplicate documents in the initial set have been processed, the final set of near duplicate documents might be set to the most recent updated set of near duplicate documents (or to the initial set of near duplicate documents in the event that the determined similarity value was never less than the threshold) (Block 470) before the method 400 is left (Node 480).
[0073] Referring back to block 410, in some embodiments consistent with the present invention, the set of documents might be Web pages. Alternatively, the set of documents might be a set of token sequence bit strings.
[0074] Referring back to block 430, in some embodiments consistent with the present invention, the second document similarity technique might include (i) projecting each of a number of token sequence bit strings into b-dimensional space by randomly choosing- a predetermined number b of entries from {-1, 11, (ii) for each document, creating a b-dimensional vector by adding the projections of all the tokens in its token sequence, and creating a final vector for the document by setting every positive entry in the b-dimensional vector to 1 and every non-positive entry to 0, and (iii) determining a similarity between two documents based on a number of bits that agree in corresponding projections of the two documents.
Referring back to block 440, in such an embodiment, the predetermined number b might be 384 and the threshold might be set to 372. In some embodiments consistent with the present invention, the threshold might be set to approximately 97% (or at least 96%) of the predetermined number b. The predetermined number b might be a lower value. For example, setting b to values as low as 192 has provided good results. Indeed, the present inventor believes that setting b to values of 100 or even slightly less might provide adequate results. In these lower settings of b, the threshold might be set to about 97% of b, or at least 96% of b.
= WO 2008/019133 PCTIUS2007/017487 4.2.2 EXEMPLARY APPARATUS
(0075] Figure 5 is block diagram of a machine 500 that may be used to perform one or more of the operations discussed above, and/or to store information generated and/or used by such operations, in a manner consistent with the present invention. The machine 500 basically includes a processor(s) 510, an input/output interface unit(s) 530, a storage device(s) 520, and a system bus or network 540 for facilitating the communication of information among the coupled elements. An input device(s) 532 and an output device(s) 534 may be coupled with the input/output interface(s) 530.
[00761 The processor(s) 510 may execute machine-executable instructions to affect one or more aspects of the present invention. At least a portion of the machine executable instructions may be stored (temporarily or more permanently) on the storage device(s) 520 and/or may be received from an external source via an input interface unit 530.
[0077] Some aspects of exemplary embodiments consistent with the present invention may be performed in the general context of computer-executable instructions, such as program modules, being executed by a personal computer. However, methods may be performed by (and data structures may be stored on) other apparatus. Program modules may include routines, programs, objects, components, data structures, etc. that perform an operation(s) or implement particular abstract data types. Moreover, those skilled in the art will appreciate that at least some aspects of the present invention may be practiced with other configurations, including hand-held devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, network computers, minicomputers, set-top boxes, mainframe computers, and the like. At least some aspects of the present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices linked through a communications network. In a distributed computing environment, program modules may be located in local and/or remote memory storage devices.
[0078] In one embodiment consistent with the present invention, the machine 500 may be one or more conventional personal computers or servers. In this case, the processing unit(s) 510 may be one or more microprocessors, the bus 540 may include a system bus that couples various system components including a system memory to the processing unit(s). The system bus 540 may be any of several types of bus structures including a memory bus or memory controller, a peripheral bus, and a local bus using any of a variety of bus architectures.
The storage devices 520 may include system memory, such as read only memory (ROM) and/or random access memory (RAM). A basic input/output system (BIOS), containing basic routines that help to transfer information between elements within the personal computer, such as during start-up, may be stored in ROM. The storage device(s) 520 may also include a hard disk drive for reading from and writing to a hard disk, a magnetic disk drive for reading from or writing to a (e.g., removable) magnetic disk, and an optical disk drive for reading from or writing to a removable (magneto-) optical disk such as a compact disk or other (magneto-) optical media.
The hard disk drive, magnetic disk drive, and (magneto-) optical disk drive may be coupled with the system bus 540 by a hard disk drive interface, a magnetic disk drive interface, and an (magneto-) optical drive interface, respectively. The drives and their associated storage media may provide nonvolatile storage of machine-readable instructions, data structures, program modules and other data for the personal computer. Although the exemplary environment described herein employs a hard disk, a removable magnetic disk and a removable optical disk, those skilled in the art will appreciate that other types of storage media (with appropriate interface devices), may be used instead of, or in addition to, the storage devices introduced above.
[0079] A user may enter commands and information into the personal computer through input devices 532, such as a keyboard and pointing device (e.g., a mouse) for example. Other input devices such as a microphone, a joystick, a game pad, a satellite dish, a scanner, or the like, may also (or alternatively) be included. These and other input devices are often connected to the processing unit(s) 510 through a serial port interface 530 coupled to the system bus 540.
Input devices may be connected by other interfaces 530, such as a parallel port, a game port or a universal serial bus (USB). However, in the context of a search facility 130, no input devices, other than those needed to accept queries, and possibly those for system administration and maintenance, are needed.
[0080] The output device(s) 534 may include a monitor or other type of display device, which may also be connected to the system bus 540 via an interface 530, such as a video adapter for example. In addition to (or instead of) the monitor, the personal computer may include other (peripheral) output devices (not shown), such as speakers and printers for example. Again, in the context of a search facility 130, no output devices, other than those needed to communicate query results, and possibly those for system administration and maintenance, are needed.
[0081] The computer may operate in a networked environment which defines logical and/or physical connections to one or more remote computers, such as a remote computer. The remote 1s computer may be another personal computer, a server, a router, a network computer, a peer device or other common network node, and may include many or all of the elements described above relative to the personal computer. The logical and/or physical connections may include a local area network (LAN) and a wide area network (WAN). An intranet and the Internet may be used instead of, or in addition to, such networks.
[0082] When used in a LAN, the personal computer may be connected to the LAN
through a network interface adapter (or "NIC") 530. When used in a WAN, such as the Internet, the personal computer may include a modem or other means for establishing communications over the wide area network. In a networked environment, at least some of the program modules depicted relative to the personal computer may be stored in the remote memory storage device.
The network connections shown are exemplary and other means of establishing a communications link between the computers may be used.
[0083] Referring once again to Figure 1, the information access facility 110 maybe a personal computer, the browsing operation 112 may be an Internet browser, and the input/output interface operation(s) 118 may include communications software and hardware.
Other information access facilities 110 may be untethered devices such as mobile telephones, personal digital assistants, etc., or other information appliances such as set-top boxes, network appliances, etc.
4.2.3 REFINEMENTS, ALTERNATIVES AND EXTENSIONS
[0084] Although embodiments consistent with the present invention were described as processing Web page documents, embodiments consistent with the present invention can operate on various other types of documents (e.g., snippets extracted from other documents, text documents, spreadsheets, database records, media streams, emails, email SPAM, bit sequences, genetic sequences, nucleotide sequences, representations of chemical structures, representations of molecular structures, characteristics of physical objects, etc.). Thus, embodiments consistent with the present invention have various applications. For example, embodiments consistent with the present invention might be used for spam detection, since spam is often replicated many times, perhaps with subtle differences. As another example, embodiments consistent with the present invention might be use to detect redundant snippets of news stories.
[0085] Some near-duplicate document detection algorithms perform poorly on pairs of Web pages from the same Website. The present inventor believes that this is mostly due to =
WO 2008/019133 PCT[US2007/017487 boilerplate text. In some alternative embodiments consistent with the present invention, boilerplate might be detected, and then removed or ignored in near-duplicate document analysis.
Alternatively, or in addition, an algorithm used to analyze Web pages on the same Website to find near-duplicate documents might be different (and potentially slower) than another algorithm used to analyze pairs of Web pages on different Websites.
[0086] Referring back to block 320 of Figure 3, some embodiments consistent with the present invention might modify the first document similarity technique such that features are weighted by frequency.
[0087] Still referring to Figure 3, although the exemplary methods consistent with the present invention described the simple case of a first document similarity technique followed by a second document similarity technique, more than two document similarity techniques might be used. Alternatively, or in addition, at least two document similarity techniques might be used, at least one of which might be applied recursively.
[0088] In some exemplary embodiments described above, a Charikar-based technique was run after a Broder-Fetterly-based technique. This is because a Charikar-based technique can be tuned to a finer degree (e.g., 372 bits of 384 bit vectors match) than a Broder-Fetterly-based technique (e.g., 2 of 6 matching supershingles). Thus, in such embodiments, the second technique can be tuned to a finer degree than the first technique. However, in alternative embodiments consistent with the present invention, other considerations might be used in determining which technique to run first. For example, in some embodiments consistent with the present invention, the second technique might take longer to run (and/or require more storage) than the first technique.
[0089] Recall that documents may be processed to generate tokens. Some embodiments consistent with the present invention might apply special processing to URLs and/or images.
For example, in some embodiments consistent with the present invention, every URL contained in the text of the page might be broken at slashes and dots, and treated like a sequence of individual terms. In some embodiments consistent with the present invention, in order to distinguish pages with different images, the URL in an IMG-tag might be considered to be a term in the page. More specifically, if the URL points to a different host, the whole URL might be considered to be a term. If, on the other hand, it points to the host of the page itself, only the filename of the URL might be used as term. Thus if a page and its images on the same host are mirrored on a different host, the URLs of the IMG-tags might generate the same tokens in the original and mirrored version. URLs can be processed using other, alternative techniques.
=
Indeed, some embodiments consistent with the present invention might ignore URLs, or simply treat URLs as a term.
[0090] Although exemplary embodiments described above might use Rabin's fingerprinting technique to generate tokens, tokens can be generated using other fingerprinting techniques (e.g., fingerprinting techniques referenced in the Hoad and Zobel paper).
[0091] Some embodiments consistent with the present invention might increase recall by, in addition to determining a "final" set of near-duplicate documents as described above, determining a second final set of near-duplicate documents using a Charikar-based technique (preferably with a high threshold, such as 97% of b, or at least 96% of b). A
union of the final set and second final set of near duplicate documents is taken to obtain a "high recall" set of near duplicate documents.
4.3 CONCLUSIONS
[0092] As can be appreciated from the foregoing, improved near-duplicate detection techniques are disclosed. These near-duplicate detection techniques performed well, particularly when analyzing Web pages from the same Website. These techniques did so without sacrificing much in the number of returned correct pairs.
Claims (23)
1. A computer-implemented method comprising:
crawling documents accessible on a network to identify a set of documents, each document in the set of documents comprising a set of token sequence bit strings;
processing the set of documents to determine a first set of near-duplicate documents using a first document similarity technique, wherein the first document similarity technique is token order dependent and token frequency independent;
processing the first set of near-duplicate documents to determine a second set of near-duplicate documents using a second document similarity technique, wherein the second document similarity technique is token order independent and token frequency dependent, and the second set of near-duplicate documents are determined based on a first threshold value;
processing the set of documents to identify a third set of near-duplicate documents using the second document similarity technique, wherein the third set of near-duplicate documents are identified based on a second threshold value greater than the first threshold value; and removing a final set of near-duplicate documents from the set of documents, and then indexing any remaining documents in the set of documents, wherein the final set of near-duplicate documents is a union of the second set of near-duplicate documents and the third set of near-duplicate documents.
crawling documents accessible on a network to identify a set of documents, each document in the set of documents comprising a set of token sequence bit strings;
processing the set of documents to determine a first set of near-duplicate documents using a first document similarity technique, wherein the first document similarity technique is token order dependent and token frequency independent;
processing the first set of near-duplicate documents to determine a second set of near-duplicate documents using a second document similarity technique, wherein the second document similarity technique is token order independent and token frequency dependent, and the second set of near-duplicate documents are determined based on a first threshold value;
processing the set of documents to identify a third set of near-duplicate documents using the second document similarity technique, wherein the third set of near-duplicate documents are identified based on a second threshold value greater than the first threshold value; and removing a final set of near-duplicate documents from the set of documents, and then indexing any remaining documents in the set of documents, wherein the final set of near-duplicate documents is a union of the second set of near-duplicate documents and the third set of near-duplicate documents.
2. The computer-implemented method of claim 1 wherein the first document similarity technique determines whether two documents are near-duplicates using representations based on a subset of the tokens of the documents, and wherein the second document similarity technique determines whether two documents are near-duplicates using representations based on all of the tokens of the documents.
3. The computer-implemented method of claim 1 wherein the first document similarity technique uses set intersection to determine whether or not documents are near-duplicates, and wherein the second document similarity technique uses random projections to determine whether or not documents are near-duplicates.
4. The computer-implemented method of claim 1 wherein the first document similarity technique includes:
fingerprinting every sub-sequence of k tokens to generate one of (A) (n - k + 1) shingles, or (B) n shingles;
applying m different random permutation functions f i for 1<=i<= m to each of the shingles to generate one of (A) n - k + 1 values, or (B) n values, for each of the random permutation functions f i;
determining, for each i, the smallest value to create an m-dimensional vector of minvalues;
reducing the m-dimensional vector of minvalues to an m'-dimensional vector of supershingles by fingerprinting non-overlapping sequences of minvalues; and concluding that two documents are near-duplicates if and only if their supershingle vectors agree in at least two supershingles.
fingerprinting every sub-sequence of k tokens to generate one of (A) (n - k + 1) shingles, or (B) n shingles;
applying m different random permutation functions f i for 1<=i<= m to each of the shingles to generate one of (A) n - k + 1 values, or (B) n values, for each of the random permutation functions f i;
determining, for each i, the smallest value to create an m-dimensional vector of minvalues;
reducing the m-dimensional vector of minvalues to an m'-dimensional vector of supershingles by fingerprinting non-overlapping sequences of minvalues; and concluding that two documents are near-duplicates if and only if their supershingle vectors agree in at least two supershingles.
5. The computer-implemented method of claim 1 wherein the first document similarity technique includes:
fingerprinting every sub-sequence of k tokens to generate one of (A) (n - k + 1) shingles, or (B) n shingles;
fingerprinting each shingle by applying m different fingerprinting functions f i for 1<= i<= m to each of the shingles to generate one of (A) n - k + 1 values, or (B) n values, for each of the m fingerprinting functions f i;
determining, for each i, the smallest value to create an m-dimensional vector of minvalues;
reducing the m-dimensional vector of minvalues to an m'-dimensional vector of supershingles by fingerprinting non-overlapping sequences of minvalues; and concluding that two documents are near-duplicates if and only if their supershingle vectors agree in at least two supershingles.
fingerprinting every sub-sequence of k tokens to generate one of (A) (n - k + 1) shingles, or (B) n shingles;
fingerprinting each shingle by applying m different fingerprinting functions f i for 1<= i<= m to each of the shingles to generate one of (A) n - k + 1 values, or (B) n values, for each of the m fingerprinting functions f i;
determining, for each i, the smallest value to create an m-dimensional vector of minvalues;
reducing the m-dimensional vector of minvalues to an m'-dimensional vector of supershingles by fingerprinting non-overlapping sequences of minvalues; and concluding that two documents are near-duplicates if and only if their supershingle vectors agree in at least two supershingles.
6. The computer-implemented method of claim 5 wherein m 84, m' = 6 and k is any value from 5 to 10.
7. The computer-implemented method of claim 6 wherein k 8.
8. The computer-implemented method of claim 1, wherein the second document similarity technique includes:
projecting each token into b-dimensional space by randomly choosing a predetermined number b of entries from {-1, 1};
for each document:
creating a b-dimensional vector for the document by adding the projections of all the tokens in a token sequence of the document;
creating a final vector for the document by setting every positive entry in the b-dimensional vector for the document to 1 and every non-positive entry to 0;
and determining a similarity between two documents based on a number of bits in which the corresponding final vectors of the two documents agree.
projecting each token into b-dimensional space by randomly choosing a predetermined number b of entries from {-1, 1};
for each document:
creating a b-dimensional vector for the document by adding the projections of all the tokens in a token sequence of the document;
creating a final vector for the document by setting every positive entry in the b-dimensional vector for the document to 1 and every non-positive entry to 0;
and determining a similarity between two documents based on a number of bits in which the corresponding final vectors of the two documents agree.
9. The computer-implemented method of claim 8 wherein b= 384.
10. The computer-implemented method of claim 8 wherein b is from 100 to 384.
11. The computer-implemented method of claim 8 wherein b is set such that a bit string of 48 bytes is stored per document.
12. The computer-implemented method of claim 1 wherein the act of processing the first set of near-duplicate documents to determine a second set of near-duplicate documents using a second document similarity technique includes:
accepting the first set of near-duplicate documents;
for each pair of near-duplicate documents in the first set;
determining a similarity value using the second document similarity technique;
if the determined similarity value is less than a threshold, then removing the current pair of near-duplicate documents from the first set to generate an updated set; and setting the second set to a most recent updated set of near-duplicate documents.
accepting the first set of near-duplicate documents;
for each pair of near-duplicate documents in the first set;
determining a similarity value using the second document similarity technique;
if the determined similarity value is less than a threshold, then removing the current pair of near-duplicate documents from the first set to generate an updated set; and setting the second set to a most recent updated set of near-duplicate documents.
13. The computer-implemented method of claim 12 wherein the second document similarity technique includes:
projecting each token into b-dimensional space by randomly choosing a predetermined number b of entries from {-1, 1};
for each document;
creating a b-dimensional vector for the document by adding the projections of all the tokens in a token sequence of the document;
creating a final vector for the document by setting every positive entry in the b-dimensional vector for the document to 1 and every non-positive entry to 0;
determining a similarity between two documents based on a number of bits in which the corresponding final vectors of the two documents agree.
projecting each token into b-dimensional space by randomly choosing a predetermined number b of entries from {-1, 1};
for each document;
creating a b-dimensional vector for the document by adding the projections of all the tokens in a token sequence of the document;
creating a final vector for the document by setting every positive entry in the b-dimensional vector for the document to 1 and every non-positive entry to 0;
determining a similarity between two documents based on a number of bits in which the corresponding final vectors of the two documents agree.
14. The computer-implemented method of claim 13 wherein the predetermined number b is 384 and wherein the threshold is set to 372.
15. The computer-implemented method of claim 13 wherein the threshold is set to approximately 97% of the predetermined number b.
16. The computer-implemented method of claim 13 wherein the threshold is set to at least 96% of the predetermined number b.
17. The computer-implemented method of claim 1 wherein each token sequence bit string was generated from a web page.
18. The computer-implemented method of claim 1 wherein the first document similarity technique requires less processing time than the second document similarity technique.
19. The computer-implemented method of claim 1 wherein the first document similarity technique requires less storage to run than the second document similarity technique.
20. The computer-implemented method of claim 1 wherein the second document similarity technique can be tuned to a finer degree than the first document similarity technique.
21. The computer-implemented method of claim 1 further comprising removing boilerplate from the accepted set of documents to generate a set of preprocessed documents, wherein the act of processing the set of documents to determine a first set of near-duplicate documents using a first document similarity technique operates on the set of preprocessed documents.
22. A machine readable medium having stored thereon machine-executable instructions which, when executed by a machine, cause the machine to perform operations comprising:
crawling documents accessible on a network to identify a set of documents, each document comprising a set of token sequence bit strings;
processing the set of documents to determine a first set of near-duplicate documents using a first document similarity technique, wherein the first document similarity technique is token order dependent and token frequency independent;
processing the first set of near-duplicate documents to determine a second set of near-duplicate documents using a second document similarity technique, wherein the second document similarity technique is token order independent and token frequency dependent, and the second set of near-duplicate documents are determined based on a first threshold value;
processing the set of documents to identify a third set of near-duplicate documents using the second document similarity technique, and wherein the third set of near-duplicate documents are identified based on a second threshold value greater than the first threshold value; and removing a final set of near-duplicate documents from the set of documents, and then indexing any remaining documents in the set of documents, wherein the final set of near-duplicate documents is a union of the second set of near-duplicate documents and the third set of near-duplicate documents.
crawling documents accessible on a network to identify a set of documents, each document comprising a set of token sequence bit strings;
processing the set of documents to determine a first set of near-duplicate documents using a first document similarity technique, wherein the first document similarity technique is token order dependent and token frequency independent;
processing the first set of near-duplicate documents to determine a second set of near-duplicate documents using a second document similarity technique, wherein the second document similarity technique is token order independent and token frequency dependent, and the second set of near-duplicate documents are determined based on a first threshold value;
processing the set of documents to identify a third set of near-duplicate documents using the second document similarity technique, and wherein the third set of near-duplicate documents are identified based on a second threshold value greater than the first threshold value; and removing a final set of near-duplicate documents from the set of documents, and then indexing any remaining documents in the set of documents, wherein the final set of near-duplicate documents is a union of the second set of near-duplicate documents and the third set of near-duplicate documents.
23. The machine readable medium of claim 22 wherein when the machine-executable instructions are executed by a machine, the act of processing the first set of near-duplicate documents to determine a second set of near-duplicate documents using a second document similarity technique includes:
accepting the first set of near-duplicate documents;
for each pair of near-duplicate documents in the first set;
determining a similarity value using the second document similarity technique;
if the determined similarity value is less than a threshold, then removing the current pair of near-duplicate documents from the first set to generate an updated set;
and setting the second set to a most recent updated set of near-duplicate documents.
accepting the first set of near-duplicate documents;
for each pair of near-duplicate documents in the first set;
determining a similarity value using the second document similarity technique;
if the determined similarity value is less than a threshold, then removing the current pair of near-duplicate documents from the first set to generate an updated set;
and setting the second set to a most recent updated set of near-duplicate documents.
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US11/499,260 | 2006-08-04 | ||
US11/499,260 US8015162B2 (en) | 2006-08-04 | 2006-08-04 | Detecting duplicate and near-duplicate files |
PCT/US2007/017487 WO2008019133A2 (en) | 2006-08-04 | 2007-08-03 | Detecting duplicate and near-duplicate files |
Publications (2)
Publication Number | Publication Date |
---|---|
CA2660202A1 CA2660202A1 (en) | 2008-02-14 |
CA2660202C true CA2660202C (en) | 2013-03-12 |
Family
ID=39033519
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA2660202A Expired - Fee Related CA2660202C (en) | 2006-08-04 | 2007-08-03 | Detecting duplicate and near-duplicate files |
Country Status (5)
Country | Link |
---|---|
US (2) | US8015162B2 (en) |
EP (1) | EP2054797A4 (en) |
CN (2) | CN102982053B (en) |
CA (1) | CA2660202C (en) |
WO (1) | WO2008019133A2 (en) |
Families Citing this family (99)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6658423B1 (en) * | 2001-01-24 | 2003-12-02 | Google, Inc. | Detecting duplicate and near-duplicate files |
US7146524B2 (en) | 2001-08-03 | 2006-12-05 | Isilon Systems, Inc. | Systems and methods for providing a distributed file system incorporating a virtual hot spare |
US7685126B2 (en) | 2001-08-03 | 2010-03-23 | Isilon Systems, Inc. | System and methods for providing a distributed file system utilizing metadata to track information about data stored throughout the system |
CN1692356B (en) | 2002-11-14 | 2014-06-04 | 易斯龙系统公司 | Systems and methods for restriping files in a distributed file system |
US7739363B1 (en) * | 2003-05-09 | 2010-06-15 | Apple Inc. | Configurable offline data store |
US8238350B2 (en) | 2004-10-29 | 2012-08-07 | Emc Corporation | Message batching with checkpoints systems and methods |
US8051425B2 (en) | 2004-10-29 | 2011-11-01 | Emc Corporation | Distributed system with asynchronous execution systems and methods |
US8055711B2 (en) * | 2004-10-29 | 2011-11-08 | Emc Corporation | Non-blocking commit protocol systems and methods |
US7797283B2 (en) | 2005-10-21 | 2010-09-14 | Isilon Systems, Inc. | Systems and methods for maintaining distributed data |
US7917474B2 (en) * | 2005-10-21 | 2011-03-29 | Isilon Systems, Inc. | Systems and methods for accessing and updating distributed data |
US7788303B2 (en) | 2005-10-21 | 2010-08-31 | Isilon Systems, Inc. | Systems and methods for distributed system scanning |
US7551572B2 (en) * | 2005-10-21 | 2009-06-23 | Isilon Systems, Inc. | Systems and methods for providing variable protection |
US7848261B2 (en) * | 2006-02-17 | 2010-12-07 | Isilon Systems, Inc. | Systems and methods for providing a quiescing protocol |
US8370455B2 (en) * | 2006-03-09 | 2013-02-05 | 24/7 Media | Systems and methods for mapping media content to web sites |
US7756898B2 (en) * | 2006-03-31 | 2010-07-13 | Isilon Systems, Inc. | Systems and methods for notifying listeners of events |
US7941420B2 (en) * | 2007-08-14 | 2011-05-10 | Yahoo! Inc. | Method for organizing structurally similar web pages from a web site |
US7676465B2 (en) * | 2006-07-05 | 2010-03-09 | Yahoo! Inc. | Techniques for clustering structurally similar web pages based on page features |
US8539056B2 (en) * | 2006-08-02 | 2013-09-17 | Emc Corporation | Systems and methods for configuring multiple network interfaces |
US7752402B2 (en) * | 2006-08-18 | 2010-07-06 | Isilon Systems, Inc. | Systems and methods for allowing incremental journaling |
US7822932B2 (en) * | 2006-08-18 | 2010-10-26 | Isilon Systems, Inc. | Systems and methods for providing nonlinear journaling |
US7899800B2 (en) * | 2006-08-18 | 2011-03-01 | Isilon Systems, Inc. | Systems and methods for providing nonlinear journaling |
US7680842B2 (en) * | 2006-08-18 | 2010-03-16 | Isilon Systems, Inc. | Systems and methods for a snapshot of data |
US7676691B2 (en) | 2006-08-18 | 2010-03-09 | Isilon Systems, Inc. | Systems and methods for providing nonlinear journaling |
US7953704B2 (en) * | 2006-08-18 | 2011-05-31 | Emc Corporation | Systems and methods for a snapshot of data |
US7680836B2 (en) * | 2006-08-18 | 2010-03-16 | Isilon Systems, Inc. | Systems and methods for a snapshot of data |
US7882071B2 (en) | 2006-08-18 | 2011-02-01 | Isilon Systems, Inc. | Systems and methods for a snapshot of data |
US7590652B2 (en) | 2006-08-18 | 2009-09-15 | Isilon Systems, Inc. | Systems and methods of reverse lookup |
US8286029B2 (en) | 2006-12-21 | 2012-10-09 | Emc Corporation | Systems and methods for managing unavailable storage devices |
US7593938B2 (en) * | 2006-12-22 | 2009-09-22 | Isilon Systems, Inc. | Systems and methods of directory entry encodings |
US8234277B2 (en) * | 2006-12-29 | 2012-07-31 | Intel Corporation | Image-based retrieval for high quality visual or acoustic rendering |
US7509448B2 (en) | 2007-01-05 | 2009-03-24 | Isilon Systems, Inc. | Systems and methods for managing semantic locks |
US8966080B2 (en) * | 2007-04-13 | 2015-02-24 | Emc Corporation | Systems and methods of managing resource utilization on a threaded computer system |
US7779048B2 (en) * | 2007-04-13 | 2010-08-17 | Isilon Systems, Inc. | Systems and methods of providing possible value ranges |
US7900015B2 (en) * | 2007-04-13 | 2011-03-01 | Isilon Systems, Inc. | Systems and methods of quota accounting |
US7698317B2 (en) * | 2007-04-20 | 2010-04-13 | Yahoo! Inc. | Techniques for detecting duplicate web pages |
US20090012984A1 (en) * | 2007-07-02 | 2009-01-08 | Equivio Ltd. | Method for Organizing Large Numbers of Documents |
US7949692B2 (en) | 2007-08-21 | 2011-05-24 | Emc Corporation | Systems and methods for portals into snapshot data |
US7966289B2 (en) * | 2007-08-21 | 2011-06-21 | Emc Corporation | Systems and methods for reading objects in a file system |
US7882068B2 (en) | 2007-08-21 | 2011-02-01 | Isilon Systems, Inc. | Systems and methods for adaptive copy on write |
US7895225B1 (en) * | 2007-12-06 | 2011-02-22 | Amazon Technologies, Inc. | Identifying potential duplicates of a document in a document corpus |
US8131751B1 (en) | 2008-01-18 | 2012-03-06 | Google Inc. | Algorithms for selecting subsequences |
US8184953B1 (en) * | 2008-02-22 | 2012-05-22 | Google Inc. | Selection of hash lookup keys for efficient retrieval |
US8239387B2 (en) * | 2008-02-22 | 2012-08-07 | Yahoo! Inc. | Structural clustering and template identification for electronic documents |
US7953709B2 (en) * | 2008-03-27 | 2011-05-31 | Emc Corporation | Systems and methods for a read only mode for a portion of a storage system |
US7984324B2 (en) | 2008-03-27 | 2011-07-19 | Emc Corporation | Systems and methods for managing stalled storage devices |
US7949636B2 (en) * | 2008-03-27 | 2011-05-24 | Emc Corporation | Systems and methods for a read only mode for a portion of a storage system |
US7870345B2 (en) | 2008-03-27 | 2011-01-11 | Isilon Systems, Inc. | Systems and methods for managing stalled storage devices |
US7962523B2 (en) * | 2008-04-11 | 2011-06-14 | Yahoo! Inc. | System and method for detecting templates of a website using hyperlink analysis |
US7930306B2 (en) * | 2008-04-30 | 2011-04-19 | Msc Intellectual Properties B.V. | System and method for near and exact de-duplication of documents |
US8086953B1 (en) * | 2008-12-19 | 2011-12-27 | Google Inc. | Identifying transient portions of web pages |
US8121991B1 (en) | 2008-12-19 | 2012-02-21 | Google Inc. | Identifying transient paths within websites |
US8862691B2 (en) * | 2008-12-22 | 2014-10-14 | Microsoft Corporation | Media aggregation and presentation |
US20100169311A1 (en) * | 2008-12-30 | 2010-07-01 | Ashwin Tengli | Approaches for the unsupervised creation of structural templates for electronic documents |
US9176883B2 (en) | 2009-04-30 | 2015-11-03 | HGST Netherlands B.V. | Storage of data reference blocks and deltas in different storage devices |
US20150010143A1 (en) * | 2009-04-30 | 2015-01-08 | HGST Netherlands B.V. | Systems and methods for signature computation in a content locality based cache |
US8180773B2 (en) * | 2009-05-27 | 2012-05-15 | International Business Machines Corporation | Detecting duplicate documents using classification |
CN101788976B (en) * | 2010-02-10 | 2012-05-09 | 北京播思软件技术有限公司 | File splitting method based on contents |
US8650195B2 (en) * | 2010-03-26 | 2014-02-11 | Palle M Pedersen | Region based information retrieval system |
US8825641B2 (en) | 2010-11-09 | 2014-09-02 | Microsoft Corporation | Measuring duplication in search results |
CN102033962B (en) * | 2010-12-31 | 2012-05-30 | 中国传媒大学 | File data replication method for quick deduplication |
US8594239B2 (en) * | 2011-02-21 | 2013-11-26 | Microsoft Corporation | Estimating document similarity using bit-strings |
WO2012145912A1 (en) * | 2011-04-28 | 2012-11-01 | Google Inc. | Presenting search results for gallery web pages |
US20120290678A1 (en) * | 2011-05-12 | 2012-11-15 | International Business Machines Corporation | Dynamic, user-driven service catalog |
US9501455B2 (en) * | 2011-06-30 | 2016-11-22 | The Boeing Company | Systems and methods for processing data |
US9407463B2 (en) * | 2011-07-11 | 2016-08-02 | Aol Inc. | Systems and methods for providing a spam database and identifying spam communications |
US8954458B2 (en) | 2011-07-11 | 2015-02-10 | Aol Inc. | Systems and methods for providing a content item database and identifying content items |
US8521769B2 (en) | 2011-07-25 | 2013-08-27 | The Boeing Company | Locating ambiguities in data |
US8484170B2 (en) * | 2011-09-19 | 2013-07-09 | International Business Machines Corporation | Scalable deduplication system with small blocks |
US20130086083A1 (en) * | 2011-09-30 | 2013-04-04 | Microsoft Corporation | Transferring ranking signals from equivalent pages |
US20130097704A1 (en) * | 2011-10-13 | 2013-04-18 | Bitdefender IPR Management Ltd. | Handling Noise in Training Data for Malware Detection |
US8843784B2 (en) | 2012-09-06 | 2014-09-23 | International Business Machines Corporation | Remapping disk drive I/O in response to media errors |
US8914668B2 (en) | 2012-09-06 | 2014-12-16 | International Business Machines Corporation | Asynchronous raid stripe writes to enable response to media errors |
US8843493B1 (en) * | 2012-09-18 | 2014-09-23 | Narus, Inc. | Document fingerprint |
US20140156624A1 (en) * | 2012-12-04 | 2014-06-05 | Microsoft Corporation | Producing, Archiving and Searching Social Content |
US9563677B2 (en) | 2012-12-11 | 2017-02-07 | Melissa Data Corp. | Systems and methods for clustered matching of records using geographic proximity |
US9111183B2 (en) | 2013-01-04 | 2015-08-18 | International Business Machines Corporation | Performing a comparison between two images which are scaled to a common resolution |
US9218701B2 (en) | 2013-05-28 | 2015-12-22 | Bank Of America Corporation | Image overlay for duplicate image detection |
US9213820B2 (en) | 2013-09-10 | 2015-12-15 | Ebay Inc. | Mobile authentication using a wearable device |
EP3161668B1 (en) | 2014-06-26 | 2020-08-05 | Google LLC | Batch-optimized render and fetch architecture |
WO2015196405A1 (en) | 2014-06-26 | 2015-12-30 | Google Inc. | Optimized browser rendering process |
RU2665920C2 (en) | 2014-06-26 | 2018-09-04 | Гугл Инк. | Optimized visualization process in browser |
US9607029B1 (en) * | 2014-12-17 | 2017-03-28 | Amazon Technologies, Inc. | Optimized mapping of documents to candidate duplicate documents in a document corpus |
CN104850574B (en) * | 2015-02-15 | 2018-07-06 | 博彦科技股份有限公司 | A kind of filtering sensitive words method of text-oriented information |
US10147107B2 (en) * | 2015-06-26 | 2018-12-04 | Microsoft Technology Licensing, Llc | Social sketches |
US10381108B2 (en) * | 2015-09-16 | 2019-08-13 | Charles Jianping Zhou | Web search and information aggregation by way of molecular network |
CN105760445A (en) * | 2016-02-03 | 2016-07-13 | 北京光年无限科技有限公司 | Junk word filtering method and system |
CN105893463B (en) * | 2016-03-23 | 2019-11-05 | 广州酷狗计算机科技有限公司 | Album input method and device |
US20180234234A1 (en) * | 2017-02-10 | 2018-08-16 | Secured FTP Hosting, LLC d/b/a SmartFile | System for describing and tracking the creation and evolution of digital files |
US10417269B2 (en) | 2017-03-13 | 2019-09-17 | Lexisnexis, A Division Of Reed Elsevier Inc. | Systems and methods for verbatim-text mining |
WO2019098732A1 (en) | 2017-11-16 | 2019-05-23 | Samsung Electronics Co., Ltd. | Method and system for management and operation over image in a computing system |
US10261784B1 (en) * | 2018-06-20 | 2019-04-16 | Terbium Labs, Inc. | Detecting copied computer code using cryptographically hashed overlapping shingles |
GB201821327D0 (en) | 2018-12-31 | 2019-02-13 | Transversal Ltd | A system and method for discriminating removing boilerplate text in documents comprising structured labelled text elements |
CN112131340B (en) * | 2019-06-25 | 2024-02-20 | 杭州萤石软件有限公司 | Character string detection method, device and storage medium |
JP7400543B2 (en) * | 2020-02-28 | 2023-12-19 | 富士フイルムビジネスイノベーション株式会社 | Information processing device and program |
CN111367563B (en) * | 2020-03-06 | 2024-02-23 | 中国银行股份有限公司 | Host version merging method and device |
US11526506B2 (en) * | 2020-05-14 | 2022-12-13 | Code42 Software, Inc. | Related file analysis |
US11726779B2 (en) * | 2021-11-03 | 2023-08-15 | Sap Se | Code simplification system |
US11797486B2 (en) | 2022-01-03 | 2023-10-24 | Bank Of America Corporation | File de-duplication for a distributed database |
CN114091428A (en) * | 2022-01-20 | 2022-02-25 | 北京搜狐互联网信息服务有限公司 | Method for determining duplication of information content, related device and computer storage medium |
Family Cites Families (34)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6850252B1 (en) | 1999-10-05 | 2005-02-01 | Steven M. Hoffberg | Intelligent electronic appliance system and method |
US5465299A (en) | 1992-12-03 | 1995-11-07 | Hitachi, Ltd. | Electronic document processing system and method of forming digital signature |
US5850490A (en) | 1993-12-22 | 1998-12-15 | Xerox Corporation | Analyzing an image of a document using alternative positionings of a class of segments |
US6505160B1 (en) | 1995-07-27 | 2003-01-07 | Digimarc Corporation | Connected audio and other media objects |
US5778395A (en) | 1995-10-23 | 1998-07-07 | Stac, Inc. | System for backing up files from disk volumes on multiple nodes of a computer network |
US5909677A (en) | 1996-06-18 | 1999-06-01 | Digital Equipment Corporation | Method for determining the resemblance of documents |
US6052693A (en) | 1996-07-02 | 2000-04-18 | Harlequin Group Plc | System for assembling large databases through information extracted from text sources |
US5745900A (en) | 1996-08-09 | 1998-04-28 | Digital Equipment Corporation | Method for indexing duplicate database records using a full-record fingerprint |
US6285999B1 (en) | 1997-01-10 | 2001-09-04 | The Board Of Trustees Of The Leland Stanford Junior University | Method for node ranking in a linked database |
US6088707A (en) | 1997-10-06 | 2000-07-11 | International Business Machines Corporation | Computer system and method of displaying update status of linked hypertext documents |
US6134532A (en) | 1997-11-14 | 2000-10-17 | Aptex Software, Inc. | System and method for optimal adaptive matching of users to most relevant entity and information in real-time |
US6119124A (en) | 1998-03-26 | 2000-09-12 | Digital Equipment Corporation | Method for clustering closely resembling data objects |
US6263348B1 (en) | 1998-07-01 | 2001-07-17 | Serena Software International, Inc. | Method and apparatus for identifying the existence of differences between two files |
US6363377B1 (en) | 1998-07-30 | 2002-03-26 | Sarnoff Corporation | Search data processor |
US6240409B1 (en) | 1998-07-31 | 2001-05-29 | The Regents Of The University Of California | Method and apparatus for detecting and summarizing document similarity within large document sets |
US6317722B1 (en) | 1998-09-18 | 2001-11-13 | Amazon.Com, Inc. | Use of electronic shopping carts to generate personal recommendations |
US6360215B1 (en) | 1998-11-03 | 2002-03-19 | Inktomi Corporation | Method and apparatus for retrieving documents based on information other than document content |
JP2000187668A (en) | 1998-12-22 | 2000-07-04 | Hitachi Ltd | Grouping method and overlap excluding method |
US6873982B1 (en) | 1999-07-16 | 2005-03-29 | International Business Machines Corporation | Ordering of database search results based on user feedback |
US6718363B1 (en) * | 1999-07-30 | 2004-04-06 | Verizon Laboratories, Inc. | Page aggregation for web sites |
US6665661B1 (en) | 2000-09-29 | 2003-12-16 | Battelle Memorial Institute | System and method for use in text analysis of documents and records |
US6978419B1 (en) | 2000-11-15 | 2005-12-20 | Justsystem Corporation | Method and apparatus for efficient identification of duplicate and near-duplicate documents and text spans using high-discriminability text fragments |
US6658423B1 (en) | 2001-01-24 | 2003-12-02 | Google, Inc. | Detecting duplicate and near-duplicate files |
US7203343B2 (en) | 2001-09-21 | 2007-04-10 | Hewlett-Packard Development Company, L.P. | System and method for determining likely identity in a biometric database |
US6871200B2 (en) * | 2002-07-11 | 2005-03-22 | Forensic Eye Ltd. | Registration and monitoring system |
US20040139072A1 (en) | 2003-01-13 | 2004-07-15 | Broder Andrei Z. | System and method for locating similar records in a database |
US20040210575A1 (en) * | 2003-04-18 | 2004-10-21 | Bean Douglas M. | Systems and methods for eliminating duplicate documents |
US8296304B2 (en) * | 2004-01-26 | 2012-10-23 | International Business Machines Corporation | Method, system, and program for handling redirects in a search engine |
CA2574554A1 (en) * | 2004-07-21 | 2006-01-26 | Equivio Ltd. | A method for determining near duplicate data objects |
US7966327B2 (en) | 2004-11-08 | 2011-06-21 | The Trustees Of Princeton University | Similarity search system with compact data structures |
US20060149820A1 (en) * | 2005-01-04 | 2006-07-06 | International Business Machines Corporation | Detecting spam e-mail using similarity calculations |
US7739314B2 (en) | 2005-08-15 | 2010-06-15 | Google Inc. | Scalable user clustering based on set similarity |
US7747614B2 (en) * | 2005-10-31 | 2010-06-29 | Yahoo! Inc. | Difference control for generating and displaying a difference result set from the result sets of a plurality of search engines |
US7472121B2 (en) * | 2005-12-15 | 2008-12-30 | International Business Machines Corporation | Document comparison using multiple similarity measures |
-
2006
- 2006-08-04 US US11/499,260 patent/US8015162B2/en active Active
-
2007
- 2007-08-03 CN CN201210158895.2A patent/CN102982053B/en active Active
- 2007-08-03 WO PCT/US2007/017487 patent/WO2008019133A2/en active Application Filing
- 2007-08-03 EP EP20070836544 patent/EP2054797A4/en not_active Withdrawn
- 2007-08-03 CA CA2660202A patent/CA2660202C/en not_active Expired - Fee Related
- 2007-08-03 CN CN2007800366340A patent/CN101523343B/en active Active
-
2011
- 2011-09-02 US US13/225,342 patent/US20120290597A1/en not_active Abandoned
Also Published As
Publication number | Publication date |
---|---|
US8015162B2 (en) | 2011-09-06 |
CN102982053B (en) | 2016-08-10 |
CN102982053A (en) | 2013-03-20 |
WO2008019133A9 (en) | 2009-04-30 |
EP2054797A2 (en) | 2009-05-06 |
US20120290597A1 (en) | 2012-11-15 |
EP2054797A4 (en) | 2013-09-04 |
US20080044016A1 (en) | 2008-02-21 |
WO2008019133A3 (en) | 2008-11-20 |
WO2008019133A2 (en) | 2008-02-14 |
CN101523343A (en) | 2009-09-02 |
CN101523343B (en) | 2012-07-04 |
CA2660202A1 (en) | 2008-02-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CA2660202C (en) | Detecting duplicate and near-duplicate files | |
US9275143B2 (en) | Detecting duplicate and near-duplicate files | |
US6615209B1 (en) | Detecting query-specific duplicate documents | |
Urvoy et al. | Tracking web spam with html style similarities | |
US7890485B2 (en) | Knowledge management tool | |
US8548972B1 (en) | Near-duplicate document detection for web crawling | |
US20030172066A1 (en) | System and method for detecting duplicate and similar documents | |
US20110307479A1 (en) | Automatic Extraction of Structured Web Content | |
US8001462B1 (en) | Updating search engine document index based on calculated age of changed portions in a document | |
Sivakumar | Effectual web content mining using noise removal from web pages | |
US8577912B1 (en) | Method and system for robust hyperlinking | |
Chen et al. | TopBlend: An efficient implementation of HtmlDiff in Java | |
Glassman et al. | Syntactic clustering of the Web | |
Upstill | Document ranking using web evidence | |
Soules et al. | Toward automatic context-based attribute assignment for semantic file systems | |
Adhiya et al. | AN EFFICIENT AND NOVEL APPROACH FOR WEB SEARCH PERSONALIZATION USING WEB USAGE MINING. | |
Broder et al. | Algorithmic aspects of information retrieval on the web | |
Desikan et al. | Link Analysis in Web Mining: Techniques and Applications | |
Pamulaparty et al. | A survey on near duplicate web pages for web crawling | |
Kumari et al. | Canonization rules for detecting different URLs | |
Chen et al. | PHAROS–Personalizing Users’ Experience in Audio-Visual Online Spaces | |
Wen | Enhancing Web Search through Web Structure Mining | |
Manolescu et al. | Web search | |
Hariharan et al. | Gain based evaluation measure for ranked web results | |
VAN STEEN | SERGUEI MANKOVSKII |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
EEER | Examination request | ||
MKLA | Lapsed |
Effective date: 20170803 |